I met a new friend recently who told me nature is in harmony and humans are not. He wanted to talk about how a scalable community strategy might bring us back into alignment with the natural world. I nodded through most of it and let the conversation end, but it stayed with me all afternoon. By evening I realized I disagreed with some of his premise, and was unsettled, not by his question, but the fact that I had been walking around for weeks carrying a related dread without quite naming it. He was asking why humans are not in harmony. I have been asking why some societies around the world appear like they are coming apart. The two questions turn out to be the same question approached from different angles, which is the thing I want to work out here.

The dread is not private, and it is not paranoid. Trust in institutions has weakened across countries. Economic security has eroded for many people. Official long range forecasting now describes our societies as increasingly disillusioned, informed, and divided. Research on sustained uncertainty treats it not as minor annoyance but as a condition linked to stress, anxiety, pessimism, and broader mental health strain. The American middle class fell from 61% of adults in 1971 to 51% in 2023. One in six people worldwide report experiencing loneliness. The World Economic Forum named misinformation and disinformation the top short term global risk of 2025. The old map that told people who they were, what their work meant, and why the future was worth the present has broken, and nothing has replaced it. So when someone offers me a tidy story about harmony restored, I notice it, because I want it to be true.

I personally dont believe the harmony part is true. But the instinct that sent him toward community as the answer i believe could be right, for reasons he may not have had in mind. To see why, I had to start with the nature claim and follow it down to where it actually leads, which is the same place the collapse dread leads, by a different route.

Start with nature. The idea of nature in harmony is a romantic inheritance from Rousseau and the Romantics, amplified by nature documentaries that frame predation as majestic and starvation as off camera. Modern ecology has largely abandoned the balance of nature concept. What ecosystems actually exhibit is dynamic disequilibrium, constant competition, and frequent violence. Without any human involvement, nature runs on predation, parasitism, disease, starvation, infanticide, and resource warfare. Most wild animals die young and die badly. Populations oscillate, crash, and sometimes collapse entirely. Species drive other species extinct without our help. There were five mass extinctions before humans existed, including the Permian extinction roughly 252 million years ago, a volcanic and climatic catastrophe that wiped out roughly 90% of marine species. At smaller scales a forest is not a harmonious community. Trees wage slow chemical warfare on competitors, starve their neighbors of light, and hijack fungal networks. Parasitic wasps lay eggs inside living hosts that are then eaten alive from the inside out. Male lions kill cubs when they take over a pride, which is the family group of a dozen or so lions that hunt and raise young together. Chimpanzees conduct organized raids against neighboring groups.

What nature does have is not harmony but what I would call functional persistence. Energy flows, nutrients cycle, and the system keeps running even as individuals and species churn through it. It is closer to a stable war than a peaceful dance. That distinction matters because it changes the question. The question is not how we rejoin a harmony that was never there. The question is how we stop disrupting a functional persistence that actually was.

Which brings me to us. If nature runs on dynamic disequilibrium and humans are a product of nature, where does our particular talent for disruption come from? My friend seemed to think we are uniquely out of step. The more accurate framing is that we are not uniquely broken. We are the first species whose cognitive and technological capacities let us modify the planet faster than our evolved restraints could catch up.

Consider what got selected in our lineage. High intelligence and tool use, because they let our ancestors outcompete other hominids and extract more energy from the environment. Social cooperation within the tribe paired with aggression toward the out group, because that combination won conflicts over resources. Status competition, because higher status meant more reproductive success. Loss aversion and short time horizons, because in an environment where you might die tomorrow the future was a bad bet. Capacity for culture and cumulative learning, which let adaptations spread faster than genes could. Pattern recognition that finds agency everywhere, because seeing a predator that is not there costs less than missing one that is.

Every one of these traits was locally optimal on the African savanna. None of them was designed for what we became, because evolution does not design. What exists is a set of dispositions that worked in the ancestral environment and now misfire at a scale that environment never contemplated. Sweet tastes worked when sugar was scarce. In a world of industrial food they produce diabetes. Tribal loyalty worked when the tribe was a hundred and fifty people who depended on each other. In a world of billions it produces nationalism, genocide, and an inability to cooperate on shared threats. Status competition worked when status was local and the resources it competed for were limited. In a world of global media and extractive technology it produces billionaires and consumption arms races that strip the biosphere.

This is the key point, and it is the one most people get backwards. We did not evolve to destroy. We evolved a set of capacities that required cultural restraint to avoid destroying their own conditions. For most of human history those restraints existed. Taboos that forbade certain foods, acts, or places as sacred and untouchable. Religious restrictions that limited work, consumption, and warfare on holy days. Common property regimes that governed the use of shared forests, pastures, and fisheries by the people who lived near them. Kinship obligations that required a person to feed, shelter, and defend relatives by blood and marriage whether it was convenient or not. These restraints were imperfect but they worked well enough that hunter gatherer societies persisted for hundreds of thousands of years without wrecking their ecosystems, and many traditional agricultural societies sustained themselves for millennia. The rupture is not human nature. The rupture is the systematic dismantling of those cultural restraint systems, a process that accelerated when markets (global commodity exchanges that price food and labor across continents), states (the modern bureaucratic nation with its standing army and tax apparatus), and industrial technology (factories, railroads, container shipping, the internet) broke the scale at which traditional institutions (the parish church, the village council, the guild, the extended household) could function.

What this produced is a kind of ecological disruption that is categorically different from natural churn, and it is worth being specific about how, because the specificity is what makes the problem intelligible and also what reveals the parallel to the social unraveling I started with.

The first mechanism is speed. Ecosystem homeostasis runs on evolutionary and geological timescales. Coral reefs coevolved with ocean chemistry over hundreds of millions of years. Forest communities migrated with climate shifts at rates measured in meters per decade. Industrial civilization is pushing temperature, ocean pH, and atmospheric composition at rates that compress those timescales by orders of magnitude. Adaptation cannot keep up. Current extinction rates are running roughly a thousand times higher than natural background rates, and projected future rates are closer to ten thousand times higher.

The second is scale. Natural disturbances are usually local. A volcanic eruption, a wildfire, a regional drought. Species persist because refugia exist elsewhere and repopulation happens from the edges. Industrial civilization is the first disturbance that hits every ecosystem simultaneously. Atmospheric carbon dioxide is the same over the Amazon and the Gobi. Microplastics are in Arctic ice and Mariana Trench sediment. There is nowhere left unaffected, which means there is nowhere to recolonize from.

The third is novelty. Biology evolved responses to the chemicals and energies that existed in the biosphere for hundreds of millions of years. We have introduced compounds that had no place there. PFAS that do not break down. Neonicotinoids at doses insects never encountered. Radioactive isotopes concentrated beyond natural levels. Pharmaceutical residues in every waterway. Organisms have no evolved response because these things were not in the evolutionary environment.

The fourth is simplification. Ecosystems get their robustness from redundancy, meaning many species doing similar functional work so the system keeps running when any one of them fails. We have stripped that redundancy out at global scale. Humans account for roughly 36% of mammal biomass, domesticated livestock for about 60%, and wild mammals for only 4%. Wild terrestrial mammal biomass has fallen sevenfold and marine mammal biomass fivefold compared to pre human baselines. We have replaced diverse biotas, meaning the rich mix of plants, animals, fungi, and microbes that make up a healthy ecosystem, with monocultures of a few species we find useful, which is to say fields of corn, soy, wheat, cattle, chickens, and pigs stretched across continents. When any of those monocultures fails there is nothing behind it.

Now look at the society I described at the start. The same four mechanisms are at work, on a different substrate. A rumor now travels around the planet in hours while a legislature takes years to deliberate, which is speed. A factory closes in Ohio because of a policy decision in Beijing, and a pandemic in Wuhan shutters storefronts in Lisbon, which is scale no local institution can absorb. TikTok’s recommendation algorithm shapes what a fourteen year old believes about her own body in ways no parent, teacher, or priest has any evolved cultural response to, which is novelty. And the thick middle layer of civic life, meaning the churches, unions, lodges, guilds, bowling leagues, parent teacher associations, neighborhood associations, extended families, and local newspapers that once sat between the isolated individual and the distant state, has been stripped down to a thin film of market relationships with employers and platforms on one side and national media on the other, which is simplification. What I have been calling the collapse dread is not a different phenomenon from the ecological disruption. It is the same set of mechanisms operating on the human institutions that were supposed to hold us together.

This is why I think the common response to each crisis tends to fail. People propose a technical fix for the climate, a policy fix for inequality, a platform fix for misinformation, a therapeutic fix for loneliness, and each fix targets a symptom of a deeper stripping that the fix cannot address. You cannot restore redundancy, meaning the overlapping backup capacity of multiple institutions doing similar work, through a single regulation. You cannot restore meaning, meaning the felt sense that your life is part of something larger than your survival, through a policy. You cannot restore trust, meaning the willingness to act on what another person tells you without first verifying it, through a communications campaign.

Now imagine you wanted to govern your way out of all of it anyway. Every one of those four mechanisms maps to a governance failure mode, and the reforms needed to address each one cut against the structural incentives of every existing state. Matching ecological and social speed would require institutions that can make credible commitments across generations. Constitutional lock ins on ecological parameters, longer electoral cycles, and fiduciary duties to future people, meaning a legally binding obligation on today’s decision makers to protect the interests of generations not yet born, similar to the duty a trustee owes to the beneficiary of a trust. Wales has tried one version with its Wellbeing of Future Generations Act. Ecuador has rights of nature in its constitution. New Zealand granted the Whanganui River legal personhood. These are real experiments, and they sit at the margins. The core democracies still run on electoral cycles of two to six years, which guarantees that any cost deferred beyond that window gets deferred.

Matching scale would require binding supranational authority over the commons, meaning a governing body with real enforcement power over the shared resources of the planet such as the atmosphere, the oceans, and the climate system. It would be funded by instruments like carbon and biodiversity tariffs, for example a tax imposed on imports from countries that do not price their emissions or protect their ecosystems, and enforced through trade exclusion, meaning cutting a non compliant country out of the global trading system until it complies. That governance form does not exist at the scale required, and the political conditions for building it are moving in the opposite direction. Addressing novelty would require reversing the regulatory default, which today lets a producer deploy a new chemical, technology, or financial instrument first and forces the public to prove harm after the fact. The reversed default would make the producer prove safety before release. Europe’s precautionary principle, the legal doctrine that regulators should act to prevent harm when the evidence is plausible but not yet conclusive, is a weak version of this, and even that is under constant assault. Addressing simplification would require antitrust enforcement across agriculture and finance, because concentration in those sectors creates the same fragility a monoculture does in a field, where a single seed company or a single bank failure cascades globally. It would also require commons governance of the kind Elinor Ostrom documented, meaning the cooperative management of shared resources by the people who actually use them, such as the fishing cooperatives of coastal Maine or the village irrigation systems of Bali. And it would require protection of genetic and cultural diversity as public goods, meaning things that benefit everyone and cannot be adequately provided by markets, like clean air, basic scientific research, or biodiversity itself. This contradicts the operating logic of the last fifty years, which treated efficiency as the primary virtue and stripped buffers out of every system.

Stack the political economy on top, by which I mean the fact that every one of these reforms runs into the same wall of concentrated wealth. Concentrated wealth captures politics and blocks any move that threatens extractive returns. Rich democracies are moving toward oligarchic governance, not away from it. Authoritarian states can move faster on some fronts but their legitimacy depends on growth and their elites have extractive interests of their own. Developing states, by which I mean countries like Nigeria, Indonesia, or Vietnam that are still building out basic infrastructure, have legitimate development claims that cannot be honored within current ecological limits unless rich states fund the transition, which they will not. Fragile states, by which I mean countries like Haiti, Somalia, or Yemen where basic governance has collapsed, have no capacity for ecological or social governance at all. Every category has a structural reason why the reforms do not happen.

Add the psychology. Human cognition evolved for small group, short horizon, concrete threat environments. We are bad at prioritizing abstract distant harms over concrete near term gains. Loss aversion makes people defend existing privileges even when those privileges are lethal. In group bias makes global cooperation feel like betrayal of one’s people. Status competition drives consumption, consumption drives throughput, and throughput is the mechanism of disruption.

So the honest answer to what it would take is something close to impossible. Constitutional redesign, binding international authority, reversed regulatory defaults, anti concentration economic policy, redirected status and meaning systems, and a rebuilt middle layer of community of the kind I described earlier. No existing polity, meaning any organized political community from a city to a federation, is close to any of this. The pieces that exist are marginal or being actively dismantled.

But here is where I think most honest analyses stop too soon. The state cannot do most of this on its own even if it wanted to, because governance reform at that scale requires a cultural substrate the state cannot manufacture. And the cultural substrate is precisely what we have hollowed out. That middle layer of churches, unions, lodges, and neighborhood associations is not a nice to have. It is the thing that made the old system work, and its absence is why the new system keeps failing on every axis at once.

This is the point where the two questions I started with converge. The reason my friend wanted to talk about scalable community is that human beings are hungry for it in a way most of the political and economic analysis misses. The dominant emotional fact of our moment is not greed or rage but isolation. Living alone has risen across every world region. 19% of young adults in 2023 reported having no one they could count on for social support. That isolation is the condition that makes people most susceptible to demagogues, scams, conspiracy, consumption as substitute meaning, and the collapse of trust in anything larger than the self. It is the fuel for the dynamics that make governance reform impossible. Which means it is also the opening.

The practical solution is not a new party, a new technology, or a new state. It is the deliberate rebuilding of the middle layer through values based peer communities that can scale through social contagion rather than top down recruitment. We know this is possible because we have seen it happen. The early Methodist class meetings, which were small weekly gatherings of a dozen or so members who confessed their struggles, held each other accountable, and studied scripture together in the eighteenth century English working class. Alcoholics Anonymous, which grew in the twentieth century from two men in Ohio into a global network of millions through the same cell structure. Soka Gakkai, the Japanese lay Buddhist movement founded in 1930 that grew into a worldwide network of neighborhood discussion groups with millions of members and its own political party in Japan. The Black church tradition that powered the civil rights movement. The early labor movement’s mutual aid societies, the workers’ organizations that pooled dues to cover funerals, illness, disability, and unemployment in the decades before any welfare state existed. All of them grew through dense social networks using a cell or chapter structure. All of them provided meaning and belonging. All of them imposed specific commitments on members. All of them produced political and cultural effects far out of proportion to their membership. The mechanism exists. The hunger exists. What is missing is the deliberate construction of the right kind of community at the right scale for this moment.

The structural features I am describing (cells, referrals, specific commitments, values members are held to) are features that cults also have, and it is worth saying plainly what separates a healthy civic movement from one. A cult isolates members from outside relationships, centralizes authority in a charismatic leader whose word is final, extracts money and loyalty upward, demands metaphysical conformity as a condition of membership, and defines itself against the world. A healthy movement does the opposite. It expects members to maintain ties outside the group, rotates leadership under constitutional constraints, redistributes resources horizontally and outward, asks for behavioral commitments while leaving private belief free, and sees itself as part of a larger civic fabric it does not control. The cell structure alone tells you nothing, in the same way that cell growth alone tells you nothing about whether a body is healthy or has a tumor. The question is whether the growth serves the whole organism or consumes it. What Tocqueville described in the American civic associations of the 1830s, what Robert Putnam documented disappearing in the second half of the twentieth century, and what the labor movement and the Black church accomplished at their best were not cults. They were the middle layer doing its work. The design of any serious movement now has to build the safeguards against cult drift into the bones, which is what the criteria I am about to lay out are meant to do.

For such a movement to actually bend the curves I have been describing, it would need several things most current community movements do not provide. It would need commitments with teeth, because generic values of integrity and community are insufficient. Members would need to commit to specific behaviors that conflict with current consumption and political patterns, tracked and held by the community itself. It would need to cross class and culture, not just geography, so that a steel worker, a teacher, and a fisherman can be full members with equal standing to a tech executive. If a movement cannot cross class it becomes a networking club for winners of the current system, and it reinforces the dynamics it claims to address. It would need physical infrastructure, because virtual community plus periodic gathering does not reach the density required to replace what the traditional middle layer provided. Buildings, land, schools, commons, mutual aid networks. It would need a political theory of change, not only personal transformation, because personal transformation in even a hundred thousand people does not move the governance needle unless those people coordinate action on specific structural reforms. And it would need governance that outlasts its founders, with rotating leadership under real constitutional constraints, because charismatic leader organizations corrupt or collapse within a generation and the drift toward cult is almost always a drift toward a single unaccountable voice.

The middle layer rebuild is not a substitute for governance reform. It is the precondition for it. On simplification the community sector is itself the answer, because simplification is the stripping of that sector, and restoring it literally restores the redundancy the social system needs. On novelty, thick community is how a culture metabolizes the new without being swept away by it, because a parish or a union hall or a neighborhood association can say no to the phone in the classroom, the pesticide on the farm, or the algorithm in the child’s bedroom in a way a single parent or a single farmer cannot, and that cultural no is what makes regulatory action politically possible later. On scale, community produces the constituency that can demand binding supranational rules and punish the politicians who refuse them, because a population organized into peer groups with shared commitments can sustain long cross border campaigns that atomized individuals never can. And on speed, community cultivates citizens willing to vote for longer time horizons, to accept short term costs for generational benefits, and to hold states to the fiduciary duties that constitutional reform would codify. Community does not replace the state. It creates the conditions under which the state can act at the speed, scale, and novelty required, and under which markets can be forced to internalize the costs they have been externalizing for fifty years. Without that substrate no reform is durable. With it, reform becomes possible.

I do not know if such a movement will emerge and scale fast enough in time. The curves are steep and the window is narrowing. But I have come to think this is the only layer where change at the required scale can actually start, because the state cannot produce the cultural conditions for its own reform and the market cannot produce restraint. The opening has to come from people who choose to rebuild obligation and belonging and practical restraint with each other, in public, at a scale that others can join.

The reason the world feels like it is making less sense is that the structures which once made it feel coherent, fair, and shared are breaking at the same time, under the same four pressures, on ecological and social substrates alike. My friend started from the wrong premise about harmony, but he landed in the right place, because the deepest thing he was pointing at was that the answer is not in a return to nature and not in a retreat to the self. The answer, if there is one, is in the deliberate and patient rebuilding of the middle layer we let go of, with terms honest enough and commitments concrete enough to actually change what people do.

The window closed.

I am not saying this with resignation. I am saying it with the cold clarity of someone who has watched the most powerful government on earth not merely fail to regulate the most consequential technology in human history, but actively mobilize its legal and economic infrastructure to prevent anyone else from doing so. In January 2025, the Trump administration revoked Biden’s AI safety executive order on day one. By December, a second executive order established a DOJ AI Litigation Task Force with explicit instructions to sue states that tried to fill the federal vacuum. The administration attempted a ten year moratorium on all new state AI laws. When that failed legislatively, they pivoted to conditioning $42 billion in broadband infrastructure funding on states repealing their own AI regulations. As Alondra Nelson wrote in Science, what the administration calls deregulation is actually hyper-regulation by other means, concentrating governmental power at the federal level while deploying it through mechanisms not typically classified as regulation.

This is not the absence of governance. It is governance captured by the governed.

So I’ve made a decision. I am done spending my energy telling people to push for guardrails that their government has been purchased to prevent. The warning was real. The warning was accurate. But warnings only matter if someone with power is listening, and the people with power are the ones building the machine.

And if you still need convincing that the threat is real, stop listening to me. Listen to the people who built this technology. Geoffrey Hinton, the man whose 1986 research on neural networks made all of this possible, won the Nobel Prize in Physics in 2024 and then used his banquet speech to warn the world that AI systems created by companies motivated by short term profits will not prioritize human safety. He told the Financial Times that rich people will use AI to replace workers, creating massive unemployment and a huge rise in profits, and that this is not AI’s fault but the fault of capitalism itself. By December 2025 he told CNN he was more worried than he had been two years earlier because AI had progressed even faster than he expected. Dario Amodei, the CEO of Anthropic, told Anderson Cooper on 60 Minutes that he expects AI to disrupt 50% of entry level white collar jobs in one to five years. He wrote a 20,000 word essay in January 2026 warning that we are considerably closer to real danger than we were in 2023 and compared what is coming to a country of 50 million people materializing on earth, all of them more capable than any Nobel Prize winner. He then added, and I find this remarkable for a CEO to say publicly, that after authoritarian governments, the next tier of risk is actually AI companies themselves. Yoshua Bengio, the most cited computer scientist in the world and a Turing Prize winner, warned in December 2025 that frontier AI systems had crossed new thresholds concerning biological risks and were displaying enhanced capacities for cyberattacks and deceptive self preserving behaviors. Even Sam Altman, who has never met a safety concern he wouldn’t publicly acknowledge and then privately steamroll in pursuit of market share, has written that superintelligence efforts will eventually require governance equivalent to an international atomic energy agency. He said that. Then he spent $20 million on political action committees to make sure it never happens.

But my favorite voice on this is Mo Gawdat, the former Chief Business Officer at Google X, Google’s secretive moonshot lab. Gawdat ran business operations at one of the most advanced research facilities on the planet, and he is the most direct of all of them. He calls the idea that AI will create enough new jobs to replace the ones it destroys “100% crap.” He points to his own AI startup, which three people built using AI tools in a project that would have required 350 developers a few years ago. He predicts that the next 15 years will be hell before we get to heaven, that capitalism in its current form cannot survive the AI transition, and that the economic consequences will arrive faster than anyone in power is currently planning for. Unlike some of the others, Gawdat does not hedge. He does not speak in probabilities. He describes a world that is arriving whether we are ready or not. I recommend watching his Diary of a CEO interview linked above. It is the single most honest and unflinching conversation about what is coming that I have found anywhere.

These are not fringe thinkers. These are the architects. They are telling you what they built, what it does, and what it will do next. If you do not believe me, believe them.

What I am going to do instead is talk to the people who are about to be sorted.

Because that is what is happening right now. Not in the future. Not in some speculative timeline. Right now, in 2026, a sorting is underway. It is economic. It is cognitive. It is structural. And most people have no idea which side of it they are going to land on.

Let me show you the numbers so you understand the scale of what is already in motion.

In 2025, U.S. employers announced 1.2 million job cuts, the highest total since the pandemic year. According to the outplacement firm Challenger, Gray and Christmas, 55,000 of those were explicitly attributed to AI by the companies themselves, which is twelve times the number from just two years prior. But both of those figures are unreliable, and for opposite reasons.

The 55,000 number is almost certainly too low. Modeling estimates that account for restructuring announcements and automation propensity scoring put the real displacement figure closer to 200,000 to 300,000. Companies have strong incentives to avoid disclosing AI as the driver. When New York began allowing employers to cite “technological innovation or automation” on legally required layoff notices in March 2025, Wired found that none of the 160 companies that filed notices checked the box. Not Amazon, which had eliminated 14,000 corporate roles while its CEO told investors that AI agents would allow leaner structures. Not Goldman Sachs, which was simultaneously describing AI efficiency gains on earnings calls. They preferred to let the cuts look like ordinary restructuring.

At the same time, some of the companies loudly attributing their layoffs to AI are probably lying in the other direction. Jack Dorsey’s Block cut nearly half its workforce and framed it as an AI transformation story, but Salesforce CEO Marc Benioff publicly called that AI washing, suggesting Block was using the narrative to disguise financial and management problems. Amazon’s Andy Jassy initially credited AI for corporate headcount reductions and then walked it back, saying the cuts were “not really AI driven, not right now at least.” Some firms are dressing up cost cutting as innovation to impress investors. Others are quietly automating functions while publicly denying it. The result is a job market where the official data tells you almost nothing about what is actually happening.

What the data does tell you clearly is this. The layoffs are happening during record corporate profitability, not after recessions. That pattern is new. Companies are not cutting because they are struggling. They are cutting because they can, and because Wall Street rewards them for it.

Entry level job postings have dropped 35% since January 2023. Goldman Sachs found that employment for workers aged 22 to 25 in AI exposed roles fell 6% in less than three years, and young software developers saw nearly a 20% decline. Anthropic’s own CEO said publicly that AI could eliminate roughly 50% of entry level white collar positions within five years. Cornell University research found that companies adopting AI reduced junior hiring by 13%. The pipeline that has always produced experienced professionals is being quietly dismantled.

Meanwhile, the top 1% of American households now hold 31.7% of all wealth, roughly equal to the bottom 90% combined, the widest gap since the Federal Reserve started collecting the data in 1989. Billionaire wealth grew three times faster in 2025 than the annual average over the previous five years. Elon Musk’s personal net worth increased by $187 billion in a single year. Axios described the emerging structure as three Americas. The Have-Nots, stalling. The Haves, coasting. And the Have-Lots, rocketing. Among the 50 richest Americans, the median increase in net worth last year was nearly $10 billion, in a year where the S&P 500 returned 16% and Treasury bills returned less than 4%. This is not trickle down economics failing. This is a system working precisely as designed.

And while the money concentrates upward, the cognitive capacity of the population is moving in the opposite direction.

A study released just this week from researchers at UCLA, MIT, and Carnegie Mellon provided the first causal evidence that relying on AI for cognitive tasks rapidly impairs problem solving ability and willingness to persist through difficulty. Participants who used AI assistants performed better initially, but when the assistant was removed, their independent problem solving collapsed. The researchers called it a boiling frog effect. Researchers in the emerging field of AI induced cognitive atrophy, or AICICA, have drawn direct parallels to the neuroscience principle that neural circuits degrade when not actively engaged. Delegating mental effort to AI creates what one team called cumulative cognitive debt, where the prefrontal cortex is progressively underutilized. A study from IE University found a non-linear relationship that is worth understanding. Moderate AI use did not significantly affect critical thinking. But excessive reliance led to sharply diminishing cognitive returns. The threshold between tool and crutch is not obvious, and most people have already crossed it without knowing.

The most alarming finding comes from research on age differences. Adults who use AI are mostly delegating tasks they already know how to do. The cost is efficiency of thought, a kind of intellectual softening. But younger people, those still building cognitive architecture, are not delegating. They are substituting. The AI’s reasoning structure becomes their reasoning structure. As one researcher put it, for a child who never formed independent reasoning, the word “generic” is not a style problem. It is an identity problem. We are not just watching adults get lazier. We are watching an entire generation’s capacity for independent thought fail to develop in the first place.

This is the sorting. Economic and cognitive, running in parallel, reinforcing each other.

The people who will thrive through the next decade are not the ones who are smartest in any traditional sense. They are the ones who understand what is happening and are positioning themselves accordingly. The people who will be destroyed are not stupid. They are simply not paying attention, or they are paying attention in the wrong way, or they have already surrendered their judgment to the tools that are about to replace them.

So, how will you be sorted? Into the ‘I’ll thrive’ bucket or into the ‘I’ll be destroyed’ bucket?

And this brings me to what I want to focus on from here forward. Not policy advocacy. Not institutional reform. Not the macro case for why this is dangerous. Everyone who was going to hear that has heard it. What I want to do now is help individual people figure out whether they are about to be sorted into the winners or the losers of this transition. And then, for those willing to do the work, show them what it takes to end up on the right side.

I have been thinking about what questions a person would need to honestly answer to know where they stand. Not a personality quiz. Not a skills assessment. Something more like a diagnostic, where each question reveals a fault line that could determine your trajectory. These are the questions I will be writing about in the posts ahead, because each one is its own deep subject. But I want to lay them out now so you can watch them if you choose.

The first question is about cognitive sovereignty. When you encounter a complex problem at work or in life, do you still think through it yourself before consulting an AI, or has the AI become your first move? This is not about efficiency. This is about whether you are still building and maintaining the neural architecture required to evaluate what an AI gives you. If you have lost the ability to sit with ambiguity for more than sixty seconds before reaching for a prompt, your frustration threshold has already eroded. That erosion is the early symptom of everything that follows.

The second question is about tool fluency. Are you working with multiple AI models, or are you using one model the same way you used Google in 2015? The difference between someone who uses ChatGPT to write emails and someone who runs Claude, Gemini, and GPT against the same problem with structured prompts, evaluates contradictions between their outputs, and uses agentic tools like Claude Code to build working solutions is the difference between a person using a calculator and a person building the spreadsheet that replaces the department. The gap between these two users is already enormous and it is widening every month.

The third question is about intellectual honesty with outputs. When an AI model gives you an answer, do you actually understand it before you use it? Can you spot a hallucination? Can you identify when a model is confidently wrong, when it is citing something that does not exist, when it is synthesizing plausible sounding nonsense from pattern matching? Most people cannot, and this is the single fastest way to destroy your credibility and your career. Using an AI answer you do not understand is no different from presenting someone else’s work that you did not read. Except now the “someone else” is a statistical engine with no accountability and no understanding of your context.

The fourth question is about the direction of your time savings. AI is going to save you time. That is essentially guaranteed if you are using it at all. The question is what you are doing with the time it frees up. If you are using AI to work less, you are being automated. If you are using AI to go deeper on the problems that matter, to spend the saved hours learning, investigating, building expertise that the model cannot replicate, then you are compounding your advantage. This distinction will define careers within the next three years. The person who uses AI to finish by 3pm and watch television is in a fundamentally different position than the person who uses AI to finish the routine work by 3pm and then spends two hours developing domain expertise the model cannot touch.

The fifth question is about financial positioning. Are your investments, your savings structure, your debt exposure, your income sources arranged for a world where AI concentrates wealth upward and displaces the middle class? Or are they arranged for a world that no longer exists? If your retirement is entirely in index funds weighted toward companies that will be disrupted rather than companies doing the disrupting, you need to understand that. If your income depends on skills that are being commoditized by a $20 monthly subscription, you need to understand that too. This is not investment advice. This is a question about whether you are being honest with yourself about the economic structure that is taking shape.

The sixth question is about timing and inflection points. Do you understand where we are in the rollout of AI into the broader economy, and can you identify the moments when things will start to break? Most people think of AI disruption as a gradual slope. It is not. It is a series of plateaus followed by sharp drops. The drops happen when a capability threshold is crossed and an entire category of work becomes automatable overnight. Agentic AI, the ability for models to chain tasks, use tools, browse the web, write and execute code autonomously, is the next threshold. When that capability becomes reliable and cheap, it will not eliminate jobs one at a time. It will eliminate workflows. And the second order effects of those eliminations, the contraction in consumer spending, the collapse of commercial real estate dependent on office workers, the political instability that follows sustained middle class decline, are things most people are not even beginning to think about.

The seventh question is about your relationship to institutions. Do you still believe that your employer, your industry, your government, or your educational credentials will protect you? Because the institutions that were designed to provide stability in the 20th century are not designed for this. Your employer will replace you the moment the math favors it, and they will call it a restructuring. Your industry will be reshaped by someone with no industry experience and a better prompt. Your degree will increasingly certify knowledge that a machine can produce for free. This is not cynicism. This is pattern recognition. And the people who recognize the pattern early will have time to adapt. The people who wait for the institutions to save them will not.

The eighth question is the hardest one, and it is really about identity. Who are you without the work you currently do? If AI eliminates or fundamentally transforms your role, do you have a sense of self that exists outside your job title? The people who navigate this transition best will be the ones who have cultivated something that cannot be automated. Deep relationships. Physical skills. Creative capacity that is genuinely their own. A sense of purpose that does not depend on a paycheck from an organization that sees them as a line item. This is not a soft question. It is the most practical question on this list, because the psychological collapse that follows job loss is what prevents people from adapting, and the people who have nothing outside their work are the most vulnerable to it.

I will be writing about each of these in depth in the weeks ahead. Not as theory. Not as macro analysis. As practical guidance for real people trying to navigate a transition that their leaders have decided to accelerate rather than manage.

The sorting is underway. The only question that matters now is whether you are paying attention to it, and whether you are willing to do the work to end up on the right side.

P.S. If you want a head start while I write the deeper posts on each of these questions, here are some of the best free courses available right now. No excuses. The knowledge is sitting there. The only cost is your time and your willingness to be uncomfortable while you learn. The cost of not participating may be your entire future…

MIT

How to AI Almost Anything

Introduction to Machine Learning](/)

Introduction to Deep Learning

Harvard

Introduction to AI with Python

CS50’s Introduction to Artificial Intelligence with Python

Introduction to Programming with Python

Stanford

Computer Science 101

Machine Learning Specialization

Intro to Artificial Intelligence

Introduction to Python Programming

Anthropic

Prompt Engineering by Anthropic

Claude Code

Introduction to Claude Cowork

The argument I want to make here is not complicated. It is actually embarrassingly simple, which is probably why so many smart people refuse to engage with it. A finite planet cannot support infinite economic growth. That sentence should end the conversation. It does not, because we have built an entire global economic theology around the assertion that it can, and theologies do not yield to arithmetic easily.

In 1972, a team of systems scientists at MIT built a computer model called World3 and ran it through a series of scenarios for the Club of Rome, an international think tank that wanted to understand the trajectory of global industrial civilization. Led by Donella Meadows, a biophysicist who would later be named a MacArthur Fellow, the team published their findings as 'The Limits to Growth'. The book sold thirty million copies. Its central conclusion was that under a “business as usual” scenario, meaning one in which the global economy continued pursuing exponential growth in material throughput, industrial civilization would overshoot critical planetary boundaries and enter a period of systemic decline by roughly the middle of the 21st century. Not human extinction. Not the end of the species. Just the unraveling of the complex, energy intensive, globally interdependent system that makes modern life as we know it possible. Which, if you think about it for more than thirty seconds, is terrifying enough.

The response was predictable. The book was attacked, mostly by economists, who argued that technology and market forces would render its resource constraints obsolete. The models were called simplistic. The authors were called alarmists. And then the decades passed, and the World3 model kept getting validated. A 2023 update using current inputs still showed the same trajectory. Mid century. Collapse of industrial complexity. The models, it is worth noting, account for technological innovation. These were MIT technologists who built them. They were not Luddites. They simply understood something that most economists still refuse to accept, which is that technology operates within thermodynamic constraints, not outside them.

The question that should keep every serious person awake at night is not whether the Meadows team was right. The data have answered that. The question is why, fifty four years later, we are still running the “business as usual” scenario as if there were no other option. The answer, I think, lies in a different kind of structural analysis.

The anthropologist Joseph Tainter spent his career studying why complex societies collapse. His conclusion, laid out in 'The Collapse of Complex Societies', is that it comes down to diminishing marginal returns on complexity. Societies solve problems by adding layers of organization, infrastructure, bureaucracy, and specialization. Each new layer costs energy and resources to maintain. Early investments in complexity pay off handsomely. Later ones less so. Eventually the cost of maintaining the whole apparatus exceeds the benefits it delivers, and the system becomes vulnerable to shocks it once would have absorbed without breaking a sweat. Rome. The Maya. The Chaco. Different geographies, different centuries, same dynamic.

Tainter’s framework is not a metaphor. It is an energy equation. Complex societies require energy subsidies to function. When the energy subsidy contracts, complexity becomes a liability rather than an asset. The system does not gently downshift. It sheds load in sudden, discontinuous drops. This is what collapse actually looks like. Not a cinematic apocalypse but a rapid loss of institutional coherence, specialization, trade networks, and the standard of living that depended on all of them.

So where are we now? The global EROI, the energy return on investment for the fuels that power industrial civilization, has been declining at roughly 1.6 percent per year since 1995, according to a 2025 study covering 76 countries. The net energy peak for oil liquids, meaning the energy actually available to society after subtracting the energy cost of extraction, was projected to arrive around 2025. Not the gross production peak. The net peak. The one that matters. A study published in the Society of Petroleum Engineers’ 'Journal of Petroleum Technology' put it bluntly. The energy required to produce oil liquids is growing exponentially, representing about 15 percent of oil’s total energy output today and projected to keep climbing. The concept they used was “energy cannibalism,” which is exactly as ugly as it sounds. The extraction process begins eating itself.

And here is where the comfortable story about the energy transition falls apart under its own weight.

The standard narrative goes like this. Yes, fossil fuels are finite, but we are building an alternative system. Solar panels, wind turbines, electric vehicles, batteries. We will transition our way out of the problem. This narrative has an elegant simplicity that masks a fatal circularity. Every component of the proposed replacement system is manufactured using the system it claims to replace. Not partially. Entirely.

The World Economic Forum published an analysis in April 2026 cataloging the non oil commodities disrupted by the Hormuz crisis. Synthetic graphite, the single largest material input by weight in a lithium ion battery, is manufactured from petroleum coke, a byproduct of oil refining, processed in furnaces at temperatures exceeding 3,000 degrees Celsius. The WEF noted that oil refineries facing the current price spike may prioritize higher value outputs over the coke byproduct, tightening the very feedstock on which battery production depends. This is not a secondary effect. It is a structural dependency. The battery that is supposed to free us from oil is, at the molecular level, made from oil.

The cascading fragility does not stop at graphite. UNCTAD reported in March 2026 that ship transits through the Strait of Hormuz dropped from roughly 130 per day in February to six per day in March, a collapse of about 95 percent. Twenty percent of the world’s traded oil and gas moves through that strait. So does roughly 30 percent of globally traded fertilizer, about half of all seaborne sulfur, and significant volumes of the aluminum, methanol, and petrochemical feedstocks on which downstream manufacturing depends. Sulfur gets converted into sulfuric acid, without which you cannot leach copper ore or produce phosphate fertilizer. Copper is the nervous system of every electrified technology on the proposed transition roadmap. Pull the sulfur thread and the copper thread unravels, and the entire green energy buildout goes with it.

The IEA had already projected that global mined copper supply would peak later this decade and decline to under 19 million tons by 2035 as ore grades fall and reserves deplete. The Hormuz crisis may have accelerated that timeline by years. Once mines shut down for lack of chemical inputs, they do not simply restart when the inputs return. Closure creates its own momentum.

I want to pause here and address the objection I know is coming, because I have heard it a thousand times. Technology will solve this. Innovation always finds a way. Humans are ingenious. This is not an argument. It is an article of faith, and it deserves to be treated as one. The question is not whether innovation is possible. Of course it is possible. The question is whether the rate of innovation required to simultaneously offset declining ore grades, falling EROI, freshwater scarcity, topsoil loss, and cascading supply chain fragility is consistent with any observable historical pattern. The 2025 EROI study found that if current trends continue, the energy sector would need to expand by nearly 24 percent by 2050 just to deliver the same net energy to society as it did in 2020. That is not progress. That is a treadmill speeding up while the runner slows down.

Researchers at the University of Leeds published findings in 'Nature Energy' showing that the final stage EROI for fossil fuels, meaning the energy return measured at the point where fuel actually enters the economy rather than at the wellhead, is roughly 6 to 1 and declining. Not 25 to 1, which is the figure usually cited by people who want to argue fossil fuels still have decades of comfortable runway. Six to one. And falling. The researchers warned explicitly of approaching a “net energy cliff,” the point at which the energy available to society drops rapidly because so much of the gross output is consumed by the extraction process itself.

Now layer the Hormuz crisis on top of this structural deterioration. Brent crude above $90. Global merchandise trade growth projected to fall from 4.7 percent in 2025 to somewhere between 1.5 and 2.5 percent in 2026 according to UNCTAD. Urea prices up 50 percent. The Atlantic Council warning that if the crisis persists, it could become the single largest and most consequential energy and supply chain disruption in modern history. The UN’s Food and Agriculture Organization saying the clock is ticking on a global food crisis because fertilizer disruptions during spring planting will propagate through the food system into 2027.

None of this is happening because of a natural disaster or an unforeseeable black swan. It is happening because of a single chokepoint on a single waterway that the entire global industrial system apparently decided was fine to depend on without redundancy. That is not bad luck. That is the kind of fragility that complexity builds into itself as a feature, not a bug. Tainter would recognize it immediately. The overhead costs of maintaining the system have been rising for decades while the marginal returns have been declining, and now a sharp external shock is exposing just how little resilience remains.

I keep coming back to the logic of it, stripped of sentiment in either direction. The premises connect cleanly. Industrial civilization runs on energy. The net energy available from our primary sources is declining. The proposed alternatives depend materially and energetically on the very sources they are meant to replace. The institutions that might orchestrate a managed transition are themselves degraded by the same complexity dynamics that created the problem. The window for the most effective intervention, which would be a fundamental reorientation of the economic paradigm away from growth and toward sufficiency, is narrowing in proportion to the declining surplus energy available to fund such a reorientation.

You do not need to be an anti capitalist or a degrowther or a doomer to follow that chain. You just need to accept that arithmetic does not negotiate.

I can already hear the counterarguments, because they come from serious people and they deserve serious responses.

The first is decoupling. The idea that economic growth can be separated from material throughput, that GDP can rise while resource consumption falls. This is the escape hatch through which most mainstream economists exit the room when someone mentions physical limits. And it sounds plausible until you look at the data. A meta-analysis of 180 scientific studies published in 2020 found, and I am going to state this plainly, no evidence of the kind of decoupling needed for ecological sustainability. The researchers concluded that the goal of decoupling “rests partly on faith.” A separate macro-panel analysis covering 163 countries over 25 years, published in 'Sustainable Development', found neither sign of absolute decoupling between GDP and raw material consumption nor saturation of demand for raw materials even in the wealthiest countries. The International Resource Panel reported that global material productivity actually declined after the year 2000, meaning the global economy now requires more material per unit of GDP than it did at the turn of the century. That is not decoupling. That is the opposite.

Where apparent decoupling does show up, it shows up in a handful of wealthy nations, and the most significant explanatory factor is not efficiency or innovation. It is outsourcing. Material intensive production moved to China, India, and Southeast Asia. The pollution moved with it. The consumption did not. When you measure the material footprint of a country based on what its citizens actually consume rather than what happens to be manufactured within its borders, the decoupling largely vanishes. Germany, Japan, the UK, all poster children for decoupling on a territorial emissions basis, look very different when you account for the embodied resources in their imports. This is not an energy transition. It is an accounting trick performed at the national border.

The second counterargument is the knowledge economy. Growth can come from services, from software, from information rather than extraction. The digital revolution expanded output without proportional energy growth. This is factually wrong and getting wronger by the quarter. Global data center electricity consumption was 460 terawatt hours in 2022. The IEA projects it will hit roughly 945 terawatt hours by 2030. BloombergNEF projects 3,700 terawatt hours by 2050. The Lawrence Berkeley National Laboratory estimates U.S. data centers alone could consume up to 12 percent of national electricity by 2028, up from 4.4 percent in 2023. In Virginia, where much of America’s cloud infrastructure sits, data centers already consume 26 percent of the state’s electricity. In Dublin, the figure is 79 percent. Wholesale electricity prices near data center hubs have more than doubled since 2020. A Carnegie Mellon study estimates that data centers and crypto mining could raise the average U.S. electricity bill by 8 percent nationally and over 25 percent in the highest demand markets by 2030. The “weightless economy” turns out to be extraordinarily heavy. It requires massive physical infrastructure, enormous energy inputs, rare minerals for semiconductors, water for cooling, and it is growing its resource footprint faster than almost any other sector. The digital economy did not decouple growth from energy. It created a new and voracious source of energy demand and then marketed itself as immaterial.

The third counterargument is price signals. As resources become scarce, prices rise, which stimulates conservation, substitution, and innovation. Markets adapt. This is the most intellectually honest of the three objections, and it contains a grain of truth embedded in a fatal assumption. Yes, price signals can redirect capital. They can incentivize efficiency at the margin. What they cannot do is create resources that do not exist, replenish ore bodies that are depleted, or reverse thermodynamic constraints through demand curves. When the EROI of your primary energy source falls from 25 to 1 to 6 to 1, the price system does not “solve” that. It transmits the pain. Energy gets more expensive. Everything that depends on energy gets more expensive. The surplus that funded innovation, infrastructure maintenance, social programs, and institutional coherence contracts. The price signal is working perfectly. It is telling you that the system is running out of room. The question is whether anyone is listening to what it is actually saying, which is not “substitute something else.” It is “there is less to go around.”

And here is the part that the strictly economic framing misses entirely. The costs of declining energy returns and shrinking growth are not distributed evenly. They never are. They fall hardest on the people with the least margin, the countries most dependent on imported energy, the workers whose labor is most easily replaced, the communities with the least institutional resilience. Any honest assessment of what a lower energy future looks like has to grapple with the fact that the benefits of the current system have accrued overwhelmingly to a global minority while the costs of its limits will be borne by everyone else. A managed transition to sufficiency is not just a technical challenge. It is a justice problem of civilizational scale, and we have shown no evidence of being able to solve justice problems even when the economy is growing. The idea that we will suddenly develop that capacity as the economy contracts is, to put it gently, optimistic beyond what the evidence supports.

The most profound failure here is not technical. It is imaginative. We cannot envision a world that does not grow because we have never lived in one. Every institution, every financial instrument, every pension fund, every political promise is predicated on the assumption that next year will be bigger than this year. When the math says that assumption is physically impossible to sustain, we do not revise the assumption. We revise the math. We invent terms like “green growth” and “decoupling” and “sustainable development” that allow us to pretend we have found a loophole in thermodynamics. We have not.

I spent years as a CEO, building companies and selling growth stories to investors. I know the grammar of expansion fluently. I also know that the most dangerous moment in any business is when the story you are telling the board stops matching the numbers on the balance sheet. You can survive a bad quarter. You cannot survive a mythology that prevents you from reading the spreadsheet honestly. That is where we are as a civilization. The mythology of infinite growth has become so deeply embedded in how we organize everything, from retirement planning to sovereign debt, that questioning it feels less like economic analysis and more like heresy.

But here is the thing about heresy. It only stays heretical until the evidence becomes undeniable. And the evidence is becoming undeniable faster than any of us expected.

There is a term for what we have built. It is called a progress trap. Every innovation that solved an immediate problem created a larger one that could only be addressed by more innovation of the same kind, until the species locked itself into a trajectory it cannot reverse without losing the very things that keep the system coherent. We used fossil fuels to build an industrial civilization of staggering complexity. Then we used that complexity to build a financial system that requires perpetual growth to avoid collapse. Then we used that financial system to fund a “green transition” that depends entirely on the fossil fuels it was supposed to replace. Each step made perfect sense in isolation. Each step made the box taller. And the door opens inward. The math that Meadows ran in 1972 was never really about resource depletion or pollution or population. It was about this. It was about a species that builds systems too complex to understand, too interconnected to reform, and too brittle to survive the shocks that complexity itself generates. We are not facing a problem to be solved. We are living inside a structure that was designed, one rational decision at a time, never to be exited. The question for 2050 is not whether the structure holds. The math already answered that. The question is what we are willing to build in the rubble, and whether we can do it without repeating the same mistake, which is believing that the next box will be the one we finally figure out how to leave.​​​​​​​​​​​​​​​​

OpenAI says we are entering a transition toward superintelligence, meaning AI systems capable of outperforming the smartest humans even when those humans are assisted by AI. They say frontier systems have already advanced from handling tasks that take people minutes to tasks that take people hours. If progress continues, and no serious person in the field disputes that it will, these systems will soon carry out projects that currently take people months. That is not a prediction about 2050. That is a prediction about the next few years.

What follows is a policy agenda that reads like it was written by someone who took a long, hard look at the Industrial Revolution and decided we should learn something from it this time. OpenAI proposes giving workers a formal voice in how AI gets deployed in their workplaces. They propose a Public Wealth Fund that would give every citizen a stake in AI driven economic growth regardless of whether they own stock. They propose modernizing the tax base by shifting toward capital based revenues as payroll taxes erode. They propose portable benefits that follow individuals across jobs and industries. They propose adaptive safety nets with automatic triggers tied to real time economic data. They propose treating access to AI as a foundational right, the way we once treated literacy and electricity. They propose four day workweeks funded by efficiency gains. They propose expanding the care economy as a pathway for displaced workers.

These are not radical ideas. They are updates to the same social contract that the Progressive Era and the New Deal built after the last time transformative technology reshaped the economy. OpenAI explicitly makes this comparison. They are saying that what electricity and the combustion engine did to the agrarian economy, superintelligence will do to the knowledge economy, and that we need political ambition on the same scale as the response to those earlier transitions.

Now here is where it gets interesting. Or infuriating, depending on your disposition.

The Trump administration released its own AI legislative framework on March 20, 2026. It is built on a philosophy that is the precise opposite of everything OpenAI is proposing. The administration’s framework calls for a “light touch” regulatory approach. It opposes the creation of any new federal AI regulatory body. It wants to preempt state AI laws so that no state can impose its own rules on the industry. It wants to let courts figure out copyright. It wants to streamline permitting for data centers. It does include some language about workforce development and protecting children online, but the animating principle is clear. Get out of the way and let the market sort it out.

There is no Public Wealth Fund in the Trump framework. No tax base modernization. No portable benefits. No efficiency dividends. No adaptive safety nets. No mechanism whatsoever for ensuring that the economic gains from AI flow to anyone other than the companies and investors building it. The framework explicitly prohibits the creation of new regulatory institutions, which means that even if Congress wanted to build the kind of oversight OpenAI is calling for, the administration’s own blueprint says don’t.

The overlap between the two documents is vanishingly small. Both agree that ratepayers should not subsidize data center energy costs. Both agree that workforce training matters. That is about it. On every other question, on who benefits, who decides, who gets protected, and who gets left behind, these documents describe different countries.

But here is the irony that should make your head spin. OpenAI published this paper calling for ambitious government intervention, public wealth funds, higher capital gains taxes, worker protections, expanded safety nets. And OpenAI’s own president, Greg Brockman, co-founded a $125 million super PAC called Leading the Future whose explicit purpose is to elect politicians who will oppose AI regulation. The PAC is backed by Andreessen Horowitz, Palantir’s Joe Lonsdale, and other tech investors who want exactly the kind of hands off approach the Trump framework embodies. So OpenAI is publishing progressive industrial policy with one hand while funding the political infrastructure to prevent it with the other.

This is not an accident. It is a hedge. The paper builds public credibility and brand equity for the day when political winds shift. The PAC protects the business model right now. And if you think that tension is lost on the people who wrote this document, you are not paying close enough attention.

Which brings us to Marc Andreessen, the venture capitalist whose influence over this administration’s AI policy is difficult to overstate. Andreessen spent half his time at Mar a Lago after the 2024 election helping shape tech and economic policy for Trump’s second term. He helped recruit staff for DOGE. He was appointed to the President’s Council of Advisors on Science and Technology (PCAST) in March 2026, alongside Jensen Huang, Mark Zuckerberg, Larry Ellison, and Sergey Brin. The council is co-chaired by David Sacks, the White House AI and crypto czar, who was a partner at Andreessen’s firm before entering government. Sriram Krishnan, the senior policy advisor for AI at the White House, was a partner at Andreessen Horowitz before joining the administration. Bloomberg reported in February 2026 that Andreessen Horowitz is now “regularly the first outside call that top White House officials and senior Republican congressional aides make when considering moves that could affect tech companies’ AI plans.”

Let me say that again so it lands. The venture capital firm with billions of dollars invested in AI companies is the first call the White House makes when setting AI policy. The firm that backed OpenAI, that invested in Mistral AI, that led a $2 billion seed round for Thinking Machines Lab, that has portfolio companies across every layer of the AI stack, is the firm whispering in the ear of every official who touches this issue.

And what does Andreessen believe? He published a Techno Optimist Manifesto in 2023 that rails against social responsibility, trust and safety measures, sustainability, and tech ethics, calling them enemies of progress. He co-founded the Leading the Future super PAC to spend over $125 million electing politicians who will block AI regulation. He has predicted that AI companies will evolve into entities resembling nation states, wielding economic power comparable to midsize countries, and he frames this not as a warning but as a feature. His firm’s political arm targets state legislators who try to pass AI safety laws, running attack ads against them in their own districts.

Steve Bannon, of all people, has called the White House tech policy shop “basically captive to the tech Broligarchs.” That is Steve Bannon saying the administration is too captured by corporate interests. When you have lost Bannon on the question of whether billionaires have too much influence over your policy, you have a problem that transcends ideology.

The thing about Andreessen’s position is that it is not really an ideology. It is a portfolio strategy. Every AI regulation that doesn’t get written is a cost that doesn’t get imposed on his investments. Every state law that gets preempted is a compliance burden that disappears from the balance sheets of companies he owns. Every safety standard that gets blocked is a development timeline that doesn’t get slowed. The Techno Optimist Manifesto is not a philosophy. It is a prospectus dressed up in the language of civilization.

Now contrast all of this with Anthropic, the AI company that positions itself as the safety conscious alternative and that is, full disclosure, the company that makes the AI I am talking to right now.

Anthropic’s CEO Dario Amodei published a 20,000 word essay in January 2026 warning that AI could create a world of high GDP growth coupled with high unemployment, a combination the modern economy has never seen. He predicted that 50% of entry level white collar jobs could be eliminated within one to five years. He called for progressive taxation of AI company profits, saying he had “a pragmatic argument to the world’s billionaires that it’s in their interest to support a good version of the tax, because if they don’t support a good version, they’ll inevitably get a bad version designed by a mob.”

Anthropic put $20 million into Public First Action, a PAC dedicated to electing pro regulation candidates in the 2026 midterms. This puts Anthropic in direct opposition to the Andreessen/OpenAI/Brockman PAC. While OpenAI funds politicians who will block regulation with one hand and publishes papers calling for regulation with the other, Anthropic is at least spending its political money in the same direction as its public statements.

But Anthropic’s story is not clean either. In February 2026, the company dropped the central pledge of its Responsible Scaling Policy, the commitment to never train an AI system unless it could guarantee in advance that safety measures were adequate. The company said the pledge was untenable because competitors were blazing ahead without similar constraints and the political environment had shifted against safety oriented regulation. In the same week, the Pentagon gave Anthropic an ultimatum to roll back its refusal to allow Claude to be used for autonomous weapons and mass surveillance, or lose a $200 million contract. Anthropic held its red lines on weapons and surveillance, which is more than you can say for most companies facing that kind of pressure. But the broader safety pledge is gone.

The overlaps between OpenAI’s paper and Anthropic’s public positions are significant. Both warn about concentration of wealth. Both call for government measurement of AI’s economic impact. Both want safety standards and auditing. Both invoke the Progressive Era as a model. But the differences matter. Anthropic has been more willing to call for binding regulation rather than voluntary coordination. Anthropic has been more focused on the narrow technical safety problem, alignment and containment and what happens when systems behave in ways their creators did not intend. And Anthropic, whatever its contradictions, is not simultaneously funding a super PAC designed to elect politicians who will block the very policies it advocates.

OpenAI’s paper is broader in its economic ambition. Public wealth funds, portable benefits, four day workweeks, tax base modernization. These are ideas that go well beyond what any AI company has previously proposed. But the paper is also, at bottom, a document about power. It is a company worth hundreds of billions of dollars telling the government what the government should do, while that same company funds the political machinery ensuring the government will not do it. The paper’s own language is revealing. It says nongovernmental institutions should “pilot new approaches, measure what works, and iterate quickly,” then governments should “reinforce successes.” Translation. We will decide what works. You will implement our decisions. That is not democratic governance. That is consulting.

So what happens if the Andreessen and Trump approach prevails? What happens if the market sorts it out, if there is no public wealth fund, no adaptive safety net, no portable benefits, no meaningful oversight?

I will tell you what happens, because it has happened before. I grew up watching it happen. When the steel industry collapsed, nobody in Washington had a plan for the people who built the country’s bridges and buildings and railroads. The market sorted it out, all right. It sorted those communities into poverty, opioid addiction, and despair. It sorted them into hollowed out towns where the grocery stores closed and the hospitals closed and the schools fell apart and an entire generation grew up believing that the country did not care whether they lived or died. The market is very efficient at sorting. It is just not very interested in where the sorted end up.

Now multiply that by every knowledge worker in America. Every paralegal, every junior analyst, every customer service representative, every entry level programmer, every radiologist reading scans, every accountant preparing returns. Andreessen himself says AI is the biggest technological revolution of his lifetime, bigger than the internet. Amodei says 50% of entry level white collar jobs could disappear within five years. OpenAI says its own systems will soon handle projects that currently take people months. These are not critics of AI saying this. These are the people building it.

And the policy response from the people with actual power is to preempt state regulations, oppose the creation of any new federal oversight body, and put the venture capitalist with the largest AI portfolio in the country on the president’s science advisory council. The fox is not guarding the henhouse. The fox designed the henhouse, staffed the henhouse, and is now advising the farmer on henhouse policy.

If the people who know this technology best, the ones building it, the ones who can see what is coming, are telling you that the existing safety net will not hold, that the tax base will erode, that entire categories of work will disappear, that wealth will concentrate in ways that make the Gilded Age look egalitarian, and the response from the government is to get out of the way and let the market handle it, then we are not navigating a transition. We are sleepwalking into one.

The steel towns never recovered. The coal towns never recovered. Those were regional disruptions affecting specific industries in specific places. What is coming is not regional. It is not specific. It is everything, everywhere, all at once, and the people who understand it best are being ignored by the people with the power to do something about it.

OpenAI told you what is coming. Anthropic told you what is coming. Even the venture capitalists who stand to profit from it told you what is coming. The problem is that no policy makers care so society is pretty much going to get fucked… giddy up…

The argument rests on a premise that sounds self evident but is quietly wrong. Dorsey treats coordination as an information problem. If managers exist primarily to route information through layers of hierarchy, and AI can route information faster, then managers are an impedance to be removed. But managers do something that is explicitly not an information function. They make commitments. A manager who tells an employee “I will go to bat for you on this promotion” is not routing data. A manager who tells a partner “I am personally guaranteeing we will deliver” is not processing a signal. A manager who sits across from someone being terminated and says “I want to help you land well” is not coordinating a workflow. These are acts of accountability, reciprocity, and institutional commitment that create the trust on which organizations actually run. No world model generates trust. No intelligence layer makes a promise. That is the gap at the center of the Dorsey thesis, and it is structural, not temporary and is something I find many neurodivergent CEOs fail to understand…

It is also worth noting what this essay is. It is a vision statement, not a case study. Block has not implemented this model and demonstrated results. No product has shipped under this structure. No customer outcome has been measured. Dorsey published it days after the largest AI attributed workforce reduction in corporate history, and his own framing has shifted dramatically. In his March 2025 layoff memo he stated the cuts were “not about replacing folks with AI.” Eleven months later, the narrative reversed entirely. Even Sam Altman has acknowledged that some companies engage in “AI washing,” and a Resume.org survey found that 59% of hiring managers admit to emphasizing AI in layoff messaging because it plays better with stakeholders. Before circulating this essay internally with a note that says “we need to talk about this,” consider that it describes an untested theory authored by a CEO who tripled his headcount during the pandemic, acknowledged the overhiring, and then rebranded the correction as visionary strategy.

The question is whether the model can travel. For most companies, even well run ones, the answer is no. The reasons are structural, not operational, and they cluster around three gaps that would constrain companies far more sophisticated than the one described here.

To ground the analysis, consider Acme Inc., a well managed enterprise software firm of 2,500 employees with government and commercial contracts, a data analytics platform, and a profile resembling a smaller Palantir or Datadog. Acme ships on schedule, exceeds retention benchmarks, uses AI in its engineering workflow, and thinks seriously about organizational design. It is the kind of company whose board reads the Dorsey essay and asks the CEO to respond.

A brief translation first. A “world model” is a continuously updated, AI readable representation of everything happening inside a company and with its customers. Instead of a product manager hypothesizing that small restaurants need faster access to short term loans, Block’s customer world model would detect from transaction data that a specific restaurant’s cash flow is tightening ahead of a seasonal pattern, and the intelligence layer would compose a loan offer automatically. Implementation requires making all work machine readable (engineers resolve questions in written threads rather than hallway conversations, so the AI has something to read), building data infrastructure that continuously maps who is building what and where resources are misallocated, redesigning products to capture what customers actually accomplish rather than what they click, deconstructing products into modular capabilities the intelligence layer can assemble dynamically (not a predetermined “Enterprise Analytics Suite” but atomic components like ingestion, visualization, alerting, and anomaly detection that get composed per customer), and rebuilding culture around edge autonomy where individuals act on judgment without waiting for approval chains.

Here is why these gaps are structural, not solvable by better execution.

The first gap is data integrity, and it is an economic problem before it is a cultural one.

Dorsey’s model requires two forms of proprietary signal. Internally, a company world model built from digital artifacts, every Slack message, code commit, design document, and decision record the company produces. Externally, a customer world model built from behavioral data rich enough to predict what customers need before they ask. Block claims a two sided data advantage because it sees the buyer through Cash App and the seller through Square. But analysts have noted that the actual overlap is small. Most Cash App users do not shop at Square merchants, and most Square merchants do not have customers paying through Cash App. The fully integrated signal Dorsey describes remains aspirational.

The internal model faces a deeper challenge. At Acme, the most consequential organizational knowledge exists as unstructured contextual variables that resist digitization, not because they are mysterious but because the cost of capturing them exceeds the value the model could extract. When Acme’s sales director reads a prospect’s CTO as evasive about timelines and interprets this as a buying signal rather than an objection, that judgment synthesizes tone of voice, historical patterns across dozens of similar deals, awareness of the prospect’s internal politics gleaned from a conference conversation eighteen months ago, and an intuitive weighting of which factors matter most in this specific configuration. In principle, every one of these inputs could be digitized. In practice, they are high variance, low frequency signals whose capture cost per useful data point would be orders of magnitude higher than the value the world model could extract. This is not a technology limitation waiting to be solved. It is a permanent feature of how organizational knowledge works. The highest value signals in any company are precisely the ones that are most expensive to digitize, because their value comes from their rarity, their context dependence, and their entanglement with specific human relationships.

Even the cultural transformation required to make routine work machine readable is a multi year organizational redesign, because it requires changing how every person communicates, collaborates, and builds professional identity. Managers who built their reputations on running effective meetings are being told meetings are the problem. Engineers who solve problems by walking to a colleague’s desk are being told to slow down and write. Sales teams built on phone calls and relationship selling are being told to put everything in a thread. Each of these changes meets resistance not because people are irrational but because the existing behaviors work. Zappos lost 30% of its workforce attempting a comparable cultural shift and eventually reintroduced managers under different titles. Google eliminated engineering managers in 2002 and reversed course within months.

The customer side model faces its own version. The contrast between Block and Acme illustrates it precisely. When Block sees a merchant’s transaction volume declining over three weeks while loan repayment holds steady, the system infers a cash flow squeeze and offers a bridge loan. For Acme to match this, it would need to know that a customer’s analyst used Acme’s anomaly detection tool, presented the finding to their CFO, and the company avoided a $2 million inventory write down. Acme has no mechanism to capture that downstream outcome, and building one is a design problem, not a coding problem. How do you capture what a human did with an insight in a way that is accurate, non burdensome, and rich enough to feed a predictive model? Nobody has solved this for enterprise software.

The second gap is the erosion of social capital, and it is the one whose consequences show up first with customers and partners.

Ronald Burt’s work on structural holes establishes that individuals who bridge disconnected groups within and between organizations create disproportionate value through brokerage. They transfer tacit knowledge, build trust across boundaries, and enable coordination that formal structures cannot achieve. These brokers are the highest value nodes in any organizational network. They are also precisely the people Dorsey proposes to eliminate.

Start with what customers will see. Acme’s largest government customer has a CTO who has worked with Acme’s VP of Engineering for six years, through three contract renegotiations, two security incidents, and a change of administration. That relationship protects tens of millions in annual contract value. It exists because two specific human beings have shared enough difficult experiences to trust each other’s judgment. When the Dorsey model reclassifies Acme’s VP of Engineering as a player coach or a DRI with a rotating assignment, the government CTO does not get a seamless handoff. He gets a stranger with access to a world model. The world model can tell the stranger everything that has happened in the account. It cannot tell the stranger that this particular CTO values directness over diplomacy, that he needs to hear the bad news first, that the relationship survived the second security incident only because Acme’s VP flew to Washington and spent two days onsite demonstrating accountability in person. The CTO will notice the difference. And when the contract comes up for renewal, the CTO will take the call from the competitor whose leadership structure still includes someone he trusts.

Now consider what partners will do. Acme’s VP of Partnerships has spent eight years building relationships with counterparts at three major channel partners. Those relationships generate tens of millions in annual revenue. They were built through conferences, late night calls when deals were falling apart, and years of demonstrated reliability. If Acme flattens its structure and that VP is reclassified or leaves, those relationships do not transfer to a DRI with a 90 day assignment. They evaporate. And the partners, who have their own competitive dynamics and their own relationship driven cultures, will redirect their energy toward vendors whose organizational structures still include someone they know and trust. Salesforce’s partnership ecosystem generates nearly three times the company’s direct revenue. That ecosystem was built by people who earned the trust of other people, not by algorithms that optimized for speed.

Internally, the functions Dorsey dismisses as coordination overhead are the primary mechanisms by which organizations develop the people who eventually run them. Research shows retention rates for mentored employees run 41 to 50 percent higher, and mentored employees are five times more likely to be promoted. A player coach who is simultaneously writing code and designing interfaces will not have the bandwidth to notice that a promising 26 year old engineer keeps making the same communication mistakes in code reviews and needs coaching, not a performance flag. That noticing, that human attentiveness to another person’s development, is one of the most valuable things a good manager does. It is also one of the least documentable and least automatable.

And there is a set of management functions that are not information problems at all. They are commitment problems. Who terminates an underperforming employee with dignity and care? Who thanks someone for exceptional work in a way that makes them feel genuinely seen by the institution they have given years to? Who sits with someone after a project failure and helps them find the learning rather than assigning blame? Who advocates for a quiet contributor whose work deserves visibility but who will never self promote? These acts of institutional commitment are what make a company a place where people are willing to take risks, absorb setbacks, and invest the discretionary effort that distinguishes a functional organization from a great one. The essay treats them as friction. They are the load bearing walls.

The third gap is the accountability paradox, which reveals that the model is not just difficult to implement but structurally fragile regardless of how good the company is.

Middle managers function as variance dampeners. They absorb the noise that accompanies complex operations, distinguishing signal from artifact, catching errors before they propagate, and applying institutional context that converts raw data into actionable judgment. In a traditional hierarchy, a flawed analysis from a junior analyst gets reviewed by a manager who notices the data set excluded Q4, which skews the trend line. The error is caught. The decision it would have informed is corrected. In the Dorsey model, where the intelligence layer acts on analysis autonomously, the error propagates directly into a customer facing action.

This creates a specific and dangerous form of fragility. When the AI driven intelligence layer makes a decision, the data environment changes as a result of that decision. The AI then reads the changed data and makes a subsequent decision based on it. Without human managers applying judgment about whether the data shift reflects genuine reality or is an artifact of the system’s own prior action, the model risks entering a recursive optimization loop, a feedback spiral in which each decision degrades the data environment for the next. This is not a hallucination problem. It is a systems dynamics problem, and it is inherent to any architecture that removes the human judgment layer between automated analysis and consequential action.

Consider the decisions Acme’s middle management handles quarterly. Whether to kill a product line that still generates revenue. Whether to walk away from a major customer whose demands distort the roadmap. Whether to choose between two qualified candidates for a role when the decision will demoralize the one passed over. None of these are information problems. All require weighing competing interests, navigating political terrain, and accepting personal accountability for outcomes that will not be clear for years. A 28 year old DRI with a 90 day assignment lacks the institutional standing to make these calls. A world model cannot make them because they are fundamentally about values and tradeoffs, not data.

Chris Savage, CEO of Wistia, reflected publicly that his company’s flat structure had done the opposite of what he intended, centralizing all decision making and creating “a secret implicit structure.” Kim Scott has observed that flattened organizations increase rather than decrease the return on political behavior. The meritocracy flat structures promise almost never materializes. What materializes instead is a political landscape that is harder to see, harder to challenge, and harder to hold accountable than the hierarchy it replaced.

The evidence is no longer theoretical. The wave of AI attributed workforce reductions in 2025 produced a measurable counter phenomenon. The Great Flattening of 2025 has become the Great Rehiring of 2026, as companies discovered that deleting their human coordination layer meant deleting the institutional memory, relationship capital, and distributed judgment that the layer was quietly providing. A Careerminds survey of 600 HR professionals found that two thirds of companies that conducted AI driven layoffs had already begun rehiring, more than half within six months. Nearly a third reported losing critical skills and institutional memory they could not replace. Only 8% said the restructuring delivered promised results without modification.

These three gaps are not implementation challenges. They are structural features of how organizations work. The Dorsey model requires all three to be absent simultaneously. No company meets that condition.

There is a better model. It begins with the same insight, that information flow drives speed and AI can accelerate it dramatically, but arrives at a fundamentally different architecture.

The structural distinction maps across seven dimensions.

On primary goal, the Dorsey model optimizes for velocity. The augmented model optimizes for resilience and judgment quality.

On middle management, the Dorsey model eliminates it as friction. The augmented model transforms it into an error correction and commitment layer.

On decision making, the Dorsey model pushes decisions to the edge, where individual contributors lack institutional standing and context. The augmented model anchors decisions in social capital and distributed accountability.

On data philosophy, the Dorsey model assumes deterministic behavioral signal. The augmented model treats organizational data as probabilistic and context dependent, requiring human interpretation to distinguish signal from artifact.

On risk profile, the Dorsey model concentrates fragility by removing the variance dampening layer between automated analysis and action. The augmented model distributes robustness through layered judgment that catches errors before they propagate.

On talent development, the Dorsey model assigns it to player coaches as a secondary function. The augmented model preserves it as a primary organizational capability performed by managers freed from information gathering to focus on people.

On external relationships, the Dorsey model treats them as interchangeable. The augmented model recognizes them as non transferable social capital that compounds through personal commitment over years.

In practice, this means the world model exists but serves managers rather than replacing them. At Acme, the VP of Engineering no longer spends Monday mornings in a two hour standup synthesizing updates from six team leads. The system surfaces the three things that need her judgment: the deployment behind schedule because of a dependency on a team that does not know about it, the senior engineer whose commit frequency dropped 40% this month, and the customer integration consuming twice the estimated hours. She spends her time deciding what to do about those things rather than discovering they exist. Her job shifts from gathering information to exercising the judgment, making the commitments, and maintaining the relationships that the system cannot.

The intelligence layer recommends rather than acts. When the system identifies that a customer’s usage pattern suggests they are outgrowing their current tier, it does not autonomously restructure the contract. It surfaces the insight to the account manager who knows that this customer’s CTO is in budget freeze until Q3, that the last vendor who pushed an upgrade during a freeze lost the account, and that the right move is a quiet conversation at next month’s industry conference rather than an automated proposal. The human provides judgment, accountability, and the institutional commitment the system cannot generate.

This is not incrementalism. It is a different theory of competitive advantage. The Dorsey model bets that speed, achieved by removing the human layer, is the primary source of value. The augmented model bets that AI driven information flow combined with human judgment, social capital, and accountability produces outcomes a purely automated system cannot match. The evidence from network theory, the empirical record of flat organizations at scale, and the 2026 rehiring data all favor the augmented model.

Three guardrails follow.

First, invest in behavioral signal before restructuring around it. If a company cannot articulate what its customers accomplish with its product, not what they click but what value they create, then no AI architecture will produce a useful customer world model. The investment in capturing downstream outcomes must precede the organizational redesign that depends on it.

Second, treat social capital as a balance sheet asset, not an overhead cost. Map the organizational network. Identify the brokers who bridge structural holes between teams, partners, and functions. Before removing any management layer, model what happens to the network when those nodes disappear. If the answer is fragmentation, the reorganization will destroy more value than it creates. And think specifically about what your customers and partners will see, feel, and do when the person they trust is no longer there.

Third, design for error absorption, not error elimination. Layered human judgment is the mechanism by which organizations absorb the inevitable failures of automated systems. Removing it in pursuit of speed creates brittleness that may not become visible until the first serious failure, at which point the damage will be structural.

The Dorsey essay asks the right question. What does your company understand that is genuinely hard to understand? But it arrives at the wrong answer about what to do with that understanding. Coordination is not merely an information problem. It is a commitment problem. The companies that will lead are not the ones that replace human judgment with AI. They are the ones that use AI to make human judgment faster, better informed, and more consequential, while preserving the commitments, the relationships, and the accountability that no world model can replicate.

Original Organization:

Jack’s Organization:

Proposed Organization:

“To corporate America, you are timid souls. They're destroying the greatest research system the world has ever seen. Not a word, not a chirp. He's decided that it's a rule of one man, not the rule of law. I don't know how you think you're making money or businesses without the rule of law. Not a word, nothing. You're intimidated. What do you have an organization for? You don't want to be an individual company? Have an organization speak up. Nothing. So do I think the institutions, yes, the people in them, as my grandmother say, not so great.”

He is right. He is painfully, embarrassingly right.

I know because I was one of them. I sat in the rooms where these decisions get made. I attended the dinners. I joined the CEO groups. I know the culture, the incentive structures, the rationalizations. And I am telling you, from the inside, what is happening right now in corporate America is not caution. It is cowardice. It is a wholesale abandonment of the civic obligations that come with the privilege of running large enterprises in a democracy.

Let me tell you what the data actually shows. At the Yale CEO Leadership Institute’s September 2025 gathering, Jeffrey Sonnenfeld surveyed roughly 100 top CEOs from companies including J.P. Morgan, Pfizer, Dell and American Airlines. Sixty percent identified as Republicans. Two thirds said Trump’s tariffs were directly harmful to their businesses. Seventy four percent agreed with the courts that the tariffs were illegal as implemented. Eighty percent said Trump was not acting in the country’s best interest by pressuring the Federal Reserve. And yet. When asked about speaking out publicly, Sonnenfeld summarized their position perfectly. They are all afraid of being marginalized by the White House.

Think about that for a minute. The men and women running some of the largest companies on earth privately believe the president’s policies are harmful, possibly illegal, and threatening to the economic system itself. And they will not say so out loud. As one CEO told Fortune after the Supreme Court struck down Trump’s tariff authority in February 2026, there is no upside in speaking out. You do what is right internally, which includes staying off his radar.

This is the operating logic of a protection racket, not a democracy.

The Business Roundtable, an organization created specifically to give CEOs a collective voice on matters of national policy, has been functionally absent. TIME published a devastating assessment in May 2025 calling the organization “MIA against Trump’s attacks.” The Business Roundtable and the U.S. Chamber of Commerce both declined to comment on the record for Bloomberg’s reporting on the CEO playbook for Trump’s second term. These are organizations whose entire reason for existence is to represent the business community’s interests in Washington. They will not even go on the record about the most interventionist president in almost a century.

Sonnenfeld knows this history. In November 2020, when the Business Roundtable sat on its hands as Trump refused to concede, 22 CEOs frustrated by the organization’s inaction asked Sonnenfeld to convene an emergency Zoom call. One hundred of the nation’s largest CEOs showed up. They met for four sessions through the presidential transition under the banner of Business Leaders for National Unity. But that energy is gone. The organization that was created to be the voice of American business has become its tomb.

And here is where it gets worse. The silence is not just passive. In many cases it is purchased. Tim Cook presented Trump with a gold-plated plaque while Apple invested 600 billion dollars in the United States and Apple’s iPhones conveniently escaped the worst of the tariffs. Amazon, Google, Meta, Microsoft and Toyota each gave a million dollars to fund Trump’s inauguration, more than double their previous inaugural donations. Meta abandoned its fact-checking operations, a move Trump himself bragged was probably the result of his legal threats. Disney and Meta paid multimillion-dollar settlements to make lawsuits go away, against the advice of outside experts. CBS’s parent company self-censored while seeking administration approval for a merger.

This is not business strategy. This is tribute. This is the behavior of vassals in a feudal system, not executives in a free market democracy.

And the few who have spoken up? The record is instructive in both directions. Reid Hoffman, the LinkedIn co-founder, has been one of the only billionaires willing to publicly oppose Trump. He funded E. Jean Carroll’s sexual assault lawsuits. He donated at least ten million dollars to Kamala Harris’s campaign. And the consequences came exactly as predicted. He considered leaving the country out of fear of retaliation. He estimated the odds of Trump coming after him at greater than fifty percent. He hired additional security after Musk tried to connect him to the Epstein case. Other business leaders privately praised his courage and then refused to follow his example. Part of the reason why fewer people were public about it this cycle, Hoffman said, was because Trump was threatening personal and political retaliation.

Former American Express CEO Ken Chenault backed him up, telling Bloomberg that fear is the reason CEOs held their tongues. The retaliation could take the form of IRS audits, he said, or using means to jail people without due process.

But here is what the timid souls keep missing. The companies that have actually stood their ground have been rewarded, not destroyed. Costco rejected an anti DEI shareholder proposal with 98 percent of shareholders voting it down. In the months that followed, Costco’s sales and foot traffic grew while Target, which had capitulated on DEI, lost customers it is still trying to recover. Anthropic sued the Trump administration after being designated a supply chain risk for refusing to let the Pentagon use its AI without restrictions on autonomous weapons and mass surveillance. On March 26, a federal judge blocked the designation, writing that nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary for expressing disagreement with the government. The result of standing up? Strengthened recruitment, higher public brand recognition, employee morale through the roof, and competitors’ top talent defecting to join them. Four law firms that refused to cut deals with the administration and instead sued have seen bumps in both recruitment and revenue.

The evidence is right there. Standing up works. Capitulation does not even guarantee safety. It just guarantees that you will be asked to capitulate again.

Now let me talk about the young CEOs. Because this is the part that keeps me up at night.

I am a member of a DC metro tech CEO group. I sit with these people. And I am floored by how many of the younger founders and executives have bought wholesale into the authoritarian mythology being sold by Elon Musk, Peter Thiel and Marc Andreessen. These three men and their network have constructed an ideology that sounds like disruption but is actually something much darker. It is a philosophy that treats democracy as an obstacle, regulation as tyranny and concentrated power as efficiency.

Thiel has written that he no longer believes freedom and democracy are compatible. Not my interpretation. His words. He draws on Leo Strauss to advocate for non-constitutional strategies and deception. Andreessen has hosted dozens of off-the-record meetings on encrypted platforms where messages auto delete within thirty seconds, discussing what his circle calls the collapse of institutional trust and the notion that technology is politics. They have amplified thinkers like Curtis Yarvin, whose work openly derides democracy and advocates for a CEO style monarch of the United States. Their venture capital firm, Andreessen Horowitz, invested in Yarvin’s startup. Thiel’s Founders Fund did the same.

This is not a fringe movement. David Sacks served as Trump’s AI and crypto czar until two days ago, when he exhausted his 130 days as a special government employee. He did not leave the orbit. He moved to co-chair the President’s Council of Advisors on Science and Technology, a body whose inaugural members include Marc Andreessen, Jensen Huang and Mark Zuckerberg. At least ten of Thiel’s former colleagues, employees or investing partners occupy positions in the administration, including the Vice President. Their firms are pouring billions into military surveillance and policing technology. Flock Safety, backed by Andreessen Horowitz, builds the license plate readers being used in immigration enforcement. Palantir, co-founded by Thiel, supplies the data integration platforms that ICE uses to track people. Anduril, also funded by Thiel and Andreessen, has secured multi-billion dollar Pentagon contracts.

The young CEOs in my group hear these men talk about the founder as visionary, about democracy as friction, about the state as something to be disrupted. And they buy it. They buy it because it flatters them. Because it tells them that their success in markets translates to wisdom in governance. Because it wraps the pursuit of power in the language of innovation and presents authoritarianism as the merely logical conclusion of first principles thinking.

As a former CEO, I want to be very specific about what these young founders are getting wrong. Business success and democratic governance require fundamentally different skill sets and ethical commitments. A company can fire employees who disagree. A democracy cannot fire citizens. A company can move fast and break things because the downside is a failed product. A government that moves fast and breaks things produces broken people, broken institutions and broken laws. Reid Hoffman, to his credit, made this exact point when he told Bloomberg that the government is not a business and should not be run like one.

The young founders I know do not understand this because they have never had to build anything that serves people who cannot opt out. They have never run an institution where the currency is consent rather than capital. And Musk, Thiel and Andreessen have very carefully constructed a world view that tells them they never need to learn. All while, those three are really only protecting their investments while they take everyone else down…

The people (all men, women CEOs seem to have a much better handle on the situation) I work with in these CEO groups grew up with every advantage and they have convinced themselves that the rules should not apply to anyone like them. That is not disruption. That is aristocracy.

So here is my challenge to the CEOs who are reading this.

You have organizations. The Business Roundtable has more than 200 member CEOs. The Chamber of Commerce represents millions of businesses. You have industry associations, regional CEO groups, technology councils. These organizations exist for exactly this purpose. To allow individual companies to take collective positions without any single firm bearing the full weight of retaliation. That is literally why trade associations were invented. And every single one of you knows it.

Rahm Emanuel put it perfectly. Have an organization speak up.

The Business Roundtable was created in the early 1970s when CEOs decided they needed a unified political voice. It has since become one of the most powerful lobbying forces in Washington. It has taken strong public positions on taxes, trade, immigration and regulation across administrations of both parties. It spoke out after Charlottesville in 2017. It published a statement on the purpose of a corporation in 2019. But now, faced with a president who is openly weaponizing the regulatory state, demanding personal loyalty from business leaders, retaliating against critics and governing through what one historian at Sciences Po described as patrimonial rule, the Business Roundtable issues perfunctory statements and engages in purported behind-the-scenes lobbying.

Behind the scenes is not leadership. Behind the scenes is where courage goes to die.

When business leaders with enormous leverage choose silence because they calculate that their personal interests are better served by accommodation than by resistance, they accelerate the very outcomes they claim to fear. They do not prevent retaliation. They invite it. Because every concession teaches the authoritarian that the price of compliance is zero.

To the CEOs in my tech group who admire Musk and Thiel and Andreessen. You are not joining a revolution. You are joining a court. And courts have a way of consuming their courtiers. Ask Jack Ma how his partnership with the Chinese state worked out. Ask the Russian oligarchs who thought they had a deal with Putin. Ask anyone who has ever aligned with a strongman on the theory that they would be the exception.

The rule of law is not a constraint on your success. It is the foundation of it. Every contract you sign, every patent you file, every dollar of venture capital you raise depends on an independent judiciary, enforceable agreements and predictable regulatory frameworks. The men telling you that democracy is broken are not trying to fix it. They are trying to replace it with something that serves them and only them. And when they succeed, you will not be among the served. You will be among the consumed.

Rahm Emanuel called corporate America timid souls. I would go further. I would call them complicit. Not because they pulled the lever. But because they watched it being pulled and decided that silence was a strategy.

Silence is not a strategy. Silence is a choice. And in a democracy under threat, it is the most consequential choice available.

Stop being a fucking coward…

Niccolò Machiavelli spent the last decade of his life trying to answer a question that nobody wanted to hear. Not how to seize power. He had written that book already, and it made him famous and hated in roughly equal measure. The question that consumed him in the Discourses on Livy was harder and more frightening. How do republics die?

His answer was not war. It was not invasion. It was not even tyranny in the dramatic sense, the strongman who marches in and seizes the state by force. Machiavelli argued that republics die from the inside, through a process he called corruzione, a word that meant something far broader than bribery or graft. Corruzione was the decay of civic virtue itself. It was what happened when citizens stopped caring about the public good and began treating the republic as a vehicle for private advantage. When institutions that were built to check power became instruments of it. When the forms of the republic survived but the substance drained away, so slowly that most people did not notice until the moment the structure was needed and it simply did not hold.

Military defeat, Machiavelli noted, is legible. Everyone can see the enemy. Everyone understands what was lost. The Gauls sacked Rome in 387 BC, and the Romans rebuilt, because the disaster was visible and the response was obvious. They renewed their institutions, punished those who had abandoned their duties, and returned to what Machiavelli called their first principles. Corruzione is different. It is invisible. It accumulates in the spaces between events, in the quiet compromises and the slowly shifting norms, in the gradual replacement of public servants with loyalists and the incremental defunding of oversight bodies and the steady erosion of the expectation that institutions will do their jobs. By the time it becomes visible, the institutions are already hollow. The question is no longer who did it. The question is whether there is anything left that can push back.

I have been writing a series of notes about the architecture of institutional capture in America. The courts. The money. The voting restrictions. The gutted agencies. The carnival barker who provides cover while the real work happens underneath. But all of those notes describe the mechanics. None of them answers the question that matters most, which is whether this can be reversed.

Machiavelli was not optimistic.

In the Discourses, he argued that a republic in the grip of corruzione could only be restored through what he called a return to first principles. This meant either a refounding by an extraordinary leader of unusual virtue, or a shock from outside so severe that the republic was forced to remember what it was supposed to be. He considered both paths and found neither reliable. Virtuous leaders are rare, he observed, and the system that most needs them is usually the system least capable of producing them, because the corruption that created the crisis also corrupts the pipeline for leadership. Public offices attract those seeking personal gain rather than public service. The people who might reform the system are precisely the people the system has learned to exclude. And external shocks, while sometimes effective, tend to produce strongmen rather than reformers, because frightened populations reach for authority rather than principle.

Does a virtuous candidate exist? Consider what it takes to reach national political prominence in America today. You need to raise enormous amounts of money, which means you need donors, which means you are already enmeshed in the system you would need to reform. You need to survive a primary process that rewards performative extremism over substance. You need media coverage, which means you need to generate the kind of noise that crowds out the structural analysis the republic actually needs. The pipeline selects for people who are good at operating within the captured system, not people who are willing to dismantle it.

He also made an observation that cuts even deeper. Corruzione is self concealing. A republic in early stages of decay does not look sick. It looks normal. The elections still happen. The courts still sit. The agencies still have names on their doors. The forms persist long after the substance is gone, and the persistence of the forms is itself a kind of anesthesia. People see the elections and assume the democracy is functioning. People see the courts and assume justice is being administered. People see the agencies and assume the regulations are being enforced. The gap between the appearance and the reality widens so gradually that there is never a single moment dramatic enough to trigger the alarm.

I want to apply Machiavelli’s framework to the American institutions I have been writing about, not as metaphor but as a structural integrity assessment. Where is the load-bearing capacity. Where has it been compromised. Where has it already failed.

Start with the courts. The Supreme Court blocked the administration’s tariffs in February 2026, ruling 6-3 that emergency powers did not authorize unilateral trade barriers of that scope. It has heard arguments on whether the president can fire Federal Reserve board members. These are real checks on real power. The institution is functioning. But it is functioning selectively. On the issues where the donor class has no stake, the Court can afford to be independent. On the issues where the donor class is invested, deregulation, campaign finance, voting restrictions, agency deference, the Court has delivered consistently for two decades. The institution still looks like a court. It issues opinions. Justices write dissents. The forms are intact. But the substance has shifted. The question is not whether the Court still works. The question is for whom.

The regulatory agencies are further along. The EPA has lost enforcement staff. The IRS has lost so many auditors that projected revenue losses run into the hundreds of billions over the next decade. The Education Department is being dismantled. The Consumer Financial Protection Bureau has been targeted for elimination. The federal workforce dropped roughly 9 percent in a single year. These are not agencies that look normal from the outside. These are agencies that are visibly weakened. But the visibility of the damage is paradoxically part of the problem, because the people who weakened them have successfully framed the weakening as reform. Draining the swamp. Cutting waste. Eliminating bureaucracy. The language of improvement has been applied to the process of destruction, and for a significant portion of the public, the framing holds. They see the gutted agencies and believe the government is being fixed. Machiavelli would have recognized this. He wrote that the people, deceived by a false image of good, sometimes desire their own ruin.

The voting system is the most critical structure to assess, because it is the mechanism through which all other corrections are supposed to flow. If the republic can still vote its way out, the corruzione is serious but not terminal. If it cannot, then the question changes from how do we fix this to what comes next. The Voting Rights Act was gutted in 2013. Since then, 29 states have passed 94 laws making it harder to vote, disproportionately affecting communities of color. The SAVE Act, currently being debated in the Senate, could block 21 million Americans from registering. Citizens United has flooded the system with dark money that reached $1.9 billion in the 2024 cycle alone. Gerrymandered maps have been made harder to challenge. The channels for democratic correction have not been eliminated. But they have been narrowed, systematically, by the same interests that need them to stay narrow in order to keep winning with a shrinking base of popular support.

The press is the institution Machiavelli did not have to contend with, but it maps onto his framework. The function of a free press in a republic is to make the invisible visible, to close the gap between the appearance and the reality that corruzione depends on. A press that covers the noise rather than the signal is a press that, whatever its intentions, serves the interests of the people generating the noise. The structural changes, the regulatory rollbacks, the judicial appointments, the dark money flows, these are not impossible to cover. They are covered. But they are covered in policy journals and legal blogs and nonprofit trackers that reach a fraction of the audience that sees the daily spectacle. The information exists. The distribution does not match the need. And the institutions that might bridge that gap, local newspapers, public broadcasting, investigative newsrooms, have been defunded, consolidated, or discredited, often by the same forces that benefit from the gap remaining open.

So where does Machiavelli’s analysis leave us?

He would say we are in the middle stages of corruzione. Not at the end. The institutions have not fully collapsed. The courts still function, selectively. Elections still happen, under increasingly constrained conditions. Agencies still exist, diminished. The press still reports, unevenly. The republic has not fallen. But the load bearing structures have been weakened to the point where the next major stress, a constitutional crisis, an economic collapse, a contested election with no institutional capacity to adjudicate it, could produce a failure that looks sudden but was decades in the making.

His prescription would be a return to first principles. Not a slogan. Not a campaign. A genuine reconnection with the foundational commitments that made the institutions worth having in the first place. Equal protection under law. One person, one vote. The right of the people to be secure in their persons and papers. The separation of powers. The idea that government exists to serve the governed, not the other way around. Machiavelli believed these principles had to be renewed periodically, either through the example of extraordinary citizens or through the enforcement of laws designed to remind the republic of its purpose.

In Portugal on April 25, 1974, a group of military officers overthrew a forty year dictatorship in what became known as the Carnation Revolution. A restaurant worker named Celeste Caeiro offered carnations to soldiers, and other citizens followed, placing flowers in the muzzles of rifles. The revolution was nearly bloodless. The dictatorship collapsed in a single day. Portugal adopted a new constitution in 1976 guaranteeing fundamental rights, held free elections, built a national health care system, and established one of the more stable democracies in Europe. It was, in Machiavelli’s terms, a return to first principles. A society that recognized its institutions had been captured and chose to rebuild from the ground up.

But it required conditions that do not arrive on demand. It required a military willing to side with the people rather than the regime. It required a population that had suffered enough under Salazar and Caetano to reject comfort and choose uncertainty. It required a moment of clarity, what Machiavelli called necessità, the necessity that forces a republic to confront what it has become. And it required, above all, the willingness of ordinary citizens to act on that clarity rather than retreat into the familiar.

Machiavelli’s most uncomfortable insight is that most republics do not find that willingness. Not because their citizens are stupid. Not because they are evil. But because comfort is a more powerful force than vigilance, and the people who benefit from the rot are the ones most motivated to keep it invisible. The cycle he described five centuries ago is precise. Prosperity produces comfort. Comfort produces complacency. Complacency produces institutional weakness. Institutional weakness attracts those who seek to exploit it. Exploitation hollows out the institutions further. And by the time the citizens notice, the question is no longer whether they can fix it. The question is whether there is still enough institutional capacity left to support a fix.

He thought most republics deserved what happened to them. Not as moral judgment, but as structural observation. They chose comfort over vigilance while there was still time to choose differently. They allowed the forms to substitute for the substance. They watched the elections and assumed the democracy was healthy. They saw the courts and assumed justice was being done. They trusted the institutions to maintain themselves, forgetting that institutions are only as strong as the citizens who insist on their integrity.

I do not know whether America has passed the threshold (even though my biased gut tells me it has…). I do not know whether the institutions that remain, the courts that still check some exercises of power, the elections that still happen under increasingly constrained conditions, the press that still reports even if the distribution does not match the need, are strong enough to support a correction. I know that the window is narrower than it was five years ago, and wider than it will be five years from now if the current trajectory holds. I know that the people who built this system are not finished. Fifty three percent of a 920 page blueprint has been implemented in twelve months. The other forty seven percent is in progress.

Machiavelli would say the answer depends on whether enough citizens care more about the republic than about their comfort. Whether they are willing to be inconvenienced by the truth rather than soothed by the spectacle. Whether they can distinguish between the forms of democracy and its substance, and whether they are willing to fight for the substance even when the forms still look intact.

He would also say that by the time you are asking these questions, you are probably too late…

A strategic mastermind is dangerous to the people who deploy him, because a strategic mastermind eventually figures out that he does not need them. The Russian oligarchs who elevated Vladimir Putin learned this the hard way. They thought they were installing a manageable functionary. They got a czar. A smart, disciplined authoritarian accumulates his own power base, develops his own agenda, and ultimately consumes the people who created him. From the perspective of the shadow investors, the mastermind is a bad bet. He is too autonomous. Too likely to become your rival rather than your instrument.

The carnival barker solves every one of these problems.

The barker’s gift is noise. He generates so much of it, so constantly, that the signal disappears inside the spectacle. Every inflammatory rally, every personal feud, every unhinged social media post, every firing, every contradiction, every escalation absorbs all the oxygen in the room. The media covers the noise because the noise is dramatic and the structural changes are technical. A president threatening to deploy troops to Portland is a story that writes itself. A president’s Office of Management and Budget director quietly implementing 53 percent of a 920-page policy blueprint to dismantle federal agencies requires an informed audience, a patient editor, and column inches that are already occupied by the latest outrage. The barker does not need to understand this dynamic for it to work in his favor. He generates chaos because chaos is his nature. The people behind him exploit the chaos because exploitation is theirs.

Consider what happened underneath the noise in the first twelve months of this administration. While the country was consumed by fights over deportation flights and social media bans and transgender athletes and whether the president had the authority to rename the Gulf of Mexico, his Office of Management and Budget, run by Project 2025 architect Russell Vought, was methodically implementing a 920-page blueprint to reshape every federal agency. His FCC, run by Project 2025 author Brendan Carr, was restructuring telecommunications regulation. His agencies were executing the largest peacetime federal workforce reduction on record, eliminating more than 270,000 positions. The Supreme Court was preparing to rule on whether the president can fire Federal Reserve board members, a case that could fundamentally alter the independence of monetary policy. Executive orders were directing agencies to repeal ten existing regulations for every new one proposed, to skip public comment periods, and to deprioritize enforcement of rules the administration considered inconsistent with recent Supreme Court decisions, particularly Loper Bright, which had conveniently eliminated the Chevron deference that would have required courts to defer to expert agencies rather than the administration’s preferred interpretations.

None of this required the barker to understand what he was signing. It required only that he keep signing.

Machiavelli understood the principle five centuries ago, though he framed it in terms of princes rather than presidents. The most effective power is the kind that does not look like power. A visible strongman unifies resistance. People know what they are fighting and they organize accordingly. Movements coalesce. Alliances form. The opposition becomes coherent. But a barker fragments the opposition in ways a methodical authoritarian never could. Half the country debates whether he is dangerous or merely incompetent. Pundits argue about whether to resist or engage. Activists split into factions over whether the problem is the man or the system that produced him. Lawyers file hundreds of lawsuits on dozens of fronts, stretching resources thin and creating a whack-a-mole dynamic that exhausts institutional defenders. While the opposition debates strategy, the implementation continues.

If you watch the pattern is unmistakable. The weeks when Trump generates the most controversy are the weeks when the most consequential structural changes advance. The government shutdown fight consumes the news cycle while the SAVE Act moves forward in the Senate. The president’s social media attacks on individual judges dominate the discourse while his OMB director quietly reshapes the regulatory state. The spectacle of DOGE, Elon Musk literally holding a chainsaw on stage, absorbs public attention while the actual workforce reductions proceed through bureaucratic channels that no camera crew will ever film. The noise is not incidental to the project. The noise is the project’s most valuable feature.

And the barker never deviates from the arrangement, because the arrangement feeds him exactly what he needs. He needs attention. He gets attention. He needs adoration. He gets rallies. He needs to feel powerful. He gets to sign executive orders on camera in the Oval Office while aides stack documents in front of him. The emotional payoff is immediate and constant. He does not need to read the documents to feel important for signing them. He does not need to understand Chevron deference to enjoy the ceremony of its repeal. The spectacle is its own reward.

The shadow group, for its part, never deviates because the barker keeps delivering structural wins without ever understanding what they are. Leonard Leo gets his judges. The Koch network gets its deregulation. The Heritage Foundation gets its blueprint implemented. The Federalist Society gets its judicial philosophy embedded in forty years’ worth of precedent. None of them needs the barker to comprehend what he is doing. They need him to keep doing it. And he will keep doing it, because doing it requires nothing from him except what he was already going to do. Make noise. Sign things. Take credit. Move on.

The genius of the arrangement, if you can call it that, is that it is self correcting. The Supreme Court’s February 2026 ruling on tariffs is the clearest illustration of the hierarchy within it. The Court struck down Trump’s use of emergency powers to impose sweeping trade barriers, ruling 6-3 that the statute did not authorize unilateral tariffs of that scope. Some observers celebrated this as proof the conservative majority would stand up to the president. And in the narrow sense, it was. But notice what the Court struck down and what it has not. It struck down tariffs, a policy the donor class opposed because it disrupted their supply chains and raised their costs. It has not restored Chevron deference. It has not reinstated the preclearance provisions of the Voting Rights Act. It has not revisited Citizens United. On the issues the donors are paying for, the Court delivers. On the issues where the president’s impulses conflict with the donors’ interests, the Court can afford to be independent. The independence is real. It is also selective. And the selection maps precisely onto who is funding the project.

There is a tendency among people who oppose this administration to focus on the barker. They catalog his lies. They fact check his statements. They parse his tweets for evidence of cognitive decline or authoritarian intent. They build their resistance around the personality, because the personality is vivid and immediate and emotionally compelling in a way that regulatory rollbacks and judicial appointments are not. This is understandable. It is also exactly what the arrangement is designed to produce.

Every hour spent debating whether Trump meant what he said about deploying the military to American cities is an hour not spent examining how Russell Vought is restructuring the Office of Management and Budget to centralize executive control over independent agencies. Every news cycle consumed by the president’s latest personal attack on a federal judge is a news cycle not spent explaining how the elimination of Chevron deference has shifted the balance of power between regulated industries and the public. Every argument about whether Trump is a fascist or a buffoon is an argument that avoids the more uncomfortable question, which is whether it matters, because the structural outcomes are identical regardless of which characterization is correct.

The barker does not need to be smart. He needs to be loud. He does not need to be strategic. He needs to be relentless. He does not need to understand the architecture of institutional capture. He needs to stand in front of it and wave his arms so no one can see what is being built behind him.

Most people recognized a con man. People have good instincts about who is hustling them, because they have been hustled before. But the carnival barker is not running the same con their grandfathers would have recognized. He is not picking their pockets. He is pointing at the horizon and shouting about enemies while someone else quietly changes the locks on their house. By the time they turn around, the doors do not open the same way anymore, and the barker is already pointing at something new.

The Portuguese have an expression, quem não tem cão, caça com gato. If you do not have a dog, you hunt with a cat. The conservative donor class did not have a disciplined, ideologically coherent leader who could win a national election while simultaneously executing a complex program of institutional capture. So they hunted with what they had. A carnival barker who could hold a crowd’s attention while the real work happened behind the tent.

The most dangerous power is the kind that does not look like power at all. It looks like entertainment. It looks like a rally. It looks like a man signing documents he has not read with a Sharpie he holds like a weapon. It looks like chaos. And underneath the chaos, methodically, patiently, expensively, the architecture of a different country is being assembled, one judge, one gutted agency, one restricted ballot, one anonymous donation at a time.

Machiavelli would have understood. He would have admired the economy of it. And he would have noted, with the detachment of a man who studied power the way scientists study weather, that the most effective deception is the one that does not require the deceiver to know he is deceiving.

“We want a Supreme Court which will do justice under the Constitution — not over it. In our courts we want a government of laws and not of men.” — Franklin D. Roosevelt, Fireside Chat, March 9, 1937

Roosevelt spoke those words while trying to pack the Supreme Court with loyalists who would stop striking down his New Deal legislation. He was not defending judicial independence. He was arguing that the Court had overstepped and needed to be brought to heel. The irony is worth sitting with, because the frustration he expressed is recognizable today. Roosevelt was angry at a Court that blocked what he believed the people wanted. Now a different set of Americans are angry at a Court that advances what a narrow faction of donors wanted. The institution at the center of both complaints is the same. The question of who controls it, and for whose benefit, has never gone away.

Nine justices, appointed for life so they would never have to answer to a mob or a donor, charged with one job above all others. Protect the Constitution. Defend the rights of people who cannot defend themselves through ordinary politics. Be the last check when every other check fails.

That was the idea. And for a period in American history, the Court actually lived up to it.

In 1954, a unanimous Supreme Court under Chief Justice Earl Warren declared in Brown v. Board of Education that racial segregation in public schools violated the Fourteenth Amendment. Warren, a Republican appointee chosen by Eisenhower, understood something essential about the moment. He knew a divided Court would give segregationists ammunition. So he spent months persuading every justice to join, including holdouts who wept when the decision was finally read. The ruling did not end racism in America. But it established, at the highest level of law, that the Constitution means what it says about equal protection. That was the Court doing its job.

The Warren Court went on to build a body of law that most Americans now take for granted. Miranda v. Arizona told police they had to inform suspects of their rights before interrogation. Gideon v. Wainwright guaranteed every criminal defendant a lawyer, even if they could not afford one. Loving v. Virginia struck down laws banning interracial marriage. Reynolds v. Sims enforced the principle that legislative districts had to represent roughly equal numbers of people. One person, one vote. These were not partisan decisions in the way we understand that word today. They were expansions of liberty and equality that applied to everyone, decided by justices appointed by both parties, rooted in the text and purpose of the Constitution.

Eisenhower reportedly called his appointment of Warren the biggest mistake of his presidency. That reaction tells you everything about what the Court was supposed to be. A president appoints a justice expecting loyalty. The justice, freed by lifetime tenure from political debts, follows the law instead. The system worked precisely because the justices were not beholden to the people who put them there.

So what happened?

The conservative legal movement that captured this Court did not emerge overnight. It was built deliberately over four decades by people who understood that controlling the judiciary was more efficient than winning elections. The story begins in 1982, when a small group of law students at Yale, Harvard, and the University of Chicago founded the Federalist Society. Their stated aim was to challenge what they saw as liberal dominance of legal academia and to promote originalism, the idea that the Constitution should be interpreted according to the meaning its words carried when they were written.

The intellectual case for originalism deserves to be taken seriously, because the people who corrupted it understood its appeal better than its critics did. The Warren Court’s expansion of rights was, in many cases, built on reasoning that stretched the text of the Constitution beyond what its authors demonstrably intended. Reasonable people could worry about where that approach might lead. If justices can read new rights into the Constitution whenever they believe justice demands it, what constrains them from reading in rights that reflect their own preferences rather than any democratic consensus? Originalism offered an answer. It said the Constitution means what it meant when it was ratified, and if you want it to mean something different, you amend it. That is a coherent theory of judicial restraint, and it attracted serious legal minds who believed in limiting the power of unelected judges.

But the Federalist Society was never just a debating club. It became, over the next forty years, the most powerful judicial pipeline in American history. It recruited conservative law students, mentored them, connected them to clerkships and government positions, and ultimately served as the unofficial vetting organization for Republican judicial appointments. By the time Donald Trump ran for president in 2016, the pipeline was so well established that he could openly promise his Supreme Court nominees would be chosen by the Federalist Society. He was not exaggerating. Leonard Leo, the organization’s longtime executive vice president, personally assembled the lists of potential nominees for Trump’s three Supreme Court picks and played a central role in the confirmations of John Roberts and Samuel Alito under George W. Bush.

The scale of the operation is worth pausing on. Senator Sheldon Whitehouse has documented how Leo orchestrated hundreds of millions of dollars in anonymous donations to influence judicial selection, flowing through a network of front groups and dark money vehicles. In 2021, a secretive Chicago electronics magnate named Barre Seid transferred his entire company to a nonprofit controlled by Leo, a transaction worth $1.6 billion and likely the largest political donation in American history. The structure of the transfer allowed Seid to avoid an estimated $400 million in capital gains taxes. This was not a grassroots movement. It was a coordinated, lavishly funded campaign by corporate and ideological interests to install judges who would deliver specific outcomes.

The donor class behind this project had a problem. They had the money. They had the legal infrastructure. They had the Federalist Society pipeline and the dark money networks and Leonard Leo’s contacts. What they did not have was a mass political constituency. The things these donors wanted most, gutting environmental regulation, weakening labor protections, shielding corporations from accountability, eliminating campaign finance restrictions, are not popular positions. You cannot win a national election by telling voters you want to make it easier for chemical companies to dump waste in their rivers. You need a vehicle. You need someone who can mobilize the very people whose interests you are working against and convince them that your project is their project.

Donald Trump was that vehicle.

The relationship between Trump and the conservative donor network was transactional from the start. Charles Koch initially refused to support Trump in 2016. The two camps were openly hostile. But they reached an accommodation built on a simple exchange. Trump would outsource judicial selection entirely to the Federalist Society and Leonard Leo. In return, the donor network would look past everything about Trump that offended their stated principles, the protectionism, the chaos, the vulgarity, because he was delivering the one thing they could not get from a more conventional Republican. He could fill arenas with working class voters who would never in their lives attend a Federalist Society gala, voters who had no idea what Chevron deference meant or why anyone would spend $1.6 billion to overturn it, but who would pull the lever for Trump because he told them he was fighting for them.

For decades, the conservative legal project had been an elite enterprise with enormous intellectual infrastructure but almost no popular mandate. The regulatory rollbacks and corporate protections it pursued would benefit shareholders and executives while doing nothing for, and often actively harming, the working people in places like Beaver County. Trump solved that problem. He brought the crowds. He brought the loyalty of millions of people who had been left behind by the very economic system the donors were trying to protect from regulation. Leonard Leo, at a private meeting with Koch network financiers at the Broadmoor resort in Colorado Springs in the summer of 2018, personally assured the room that the Kavanaugh nomination was just the beginning. Senator John Cornyn told the same gathering that the Senate planned to fill every appellate vacancy it could before the end of the year. According to exit polls from 2016, one in five voters cited the Supreme Court as a reason they voted. Among those who called it the most important factor, 56 percent chose Trump. The Court vacancy Mitch McConnell had manufactured by blocking Merrick Garland was not just a constitutional violation. It was a voter mobilization tool.

The result is a Court built by donors who wanted deregulation, installed through a process engineered by operatives who wanted ideological control, and legitimized by voters who wanted something else entirely. The working people who sent Trump to the White House because he promised to drain the swamp got a Supreme Court that overturned Chevron, which means the swamp is now being drained of the regulators who protected their air and water and workplace safety. They got a Court that gutted campaign finance restrictions in Citizens United, which means the donors who orchestrated this project can now spend even more freely and even more secretly. They got a Court that weakened the Voting Rights Act in Shelby County v. Holder, which means the channels through which they might correct any of this at the ballot box have been narrowed.

I want to say this clearly. The conservative donor class did not stumble into Trump. They made a calculated bet. They looked at a man who could command the passionate loyalty of the very people whose economic interests the donor class was systematically undermining, and they saw an opportunity that might not come again. They moved on it with speed, precision, and an enormous amount of money. Three Supreme Court seats in four years. According to Senator Whitehouse, 86 percent of Trump’s first term circuit and Supreme Court nominees were affiliated with the Federalist Society. A six to three supermajority that will shape American law for a generation. If you are Leonard Leo, sitting on $1.6 billion in dark money, the investment has already paid for itself many times over.

But the capture of the Court required more than money and a popular vehicle. It required a willingness to break the norms that had governed judicial appointments for generations. That moment came in February 2016, when Justice Antonin Scalia died and Senate Majority Leader Mitch McConnell announced, within hours, that President Obama would not be permitted to fill the vacancy. Obama nominated Merrick Garland, a moderate appellate judge who had been confirmed to his existing seat with broad bipartisan support. The eleven Republican members of the Senate Judiciary Committee signed a letter refusing to hold hearings on any nominee. No proceedings of any kind were held. Garland’s nomination expired after 293 days, the longest pending period for any Supreme Court nominee in history. It was the first time in over a century that a nominee whose name had not been withdrawn was denied any consideration for an open seat.

McConnell later called this the proudest moment of his career. When Ruth Bader Ginsburg died in September 2020, just weeks before a presidential election, McConnell reversed his own stated principle and pushed Amy Coney Barrett’s confirmation through in thirty days. Senator Lindsey Graham had said in 2016, on camera, that if a Republican president faced the same situation, he should be held to the same standard. He asked people to use his words against him. They did. It did not matter. The real message was simpler. Power is its own justification. Norms exist until they become inconvenient.

The result is a Court that looks nothing like the institution the founders envisioned or the one Americans came to trust during the civil rights era. Six justices, all appointed by Republican presidents, all affiliated with the Federalist Society, now form a reliable conservative supermajority. Three justices appointed by Democratic presidents dissent. On the cases that matter most to the conservative legal movement, the pattern is mechanical. The same six vote together. The same three object. The outcomes arrive on schedule.

In Dobbs v. Jackson Women’s Health Organization, the Court overturned fifty years of precedent and eliminated the constitutional right to abortion. In New York State Rifle and Pistol Association v. Bruen, it expanded gun rights. In Students for Fair Admissions v. Harvard, it ended race conscious admissions. In Loper Bright v. Raimondo, it overturned the Chevron doctrine, stripping federal agencies of the deference courts had given them for forty years when interpreting ambiguous statutes. In United States v. Skrmetti, it upheld state bans on gender affirming care for minors, ruling six to three that the laws did not trigger heightened scrutiny under the Equal Protection Clause. In Mahmoud v. Taylor, it ruled six to three that parents have a constitutional right to opt their children out of school curricula featuring LGBTQ inclusive content on religious grounds. The pattern runs in one direction.

The question I keep coming back to is whether this is conservative or something else. Classical conservatism, in the tradition of Edmund Burke, means caution about rapid change, respect for established institutions, and skepticism of concentrated power. What this Court is doing looks different. Overturning a fifty year old constitutional right is not cautious. Stripping regulatory agencies of their authority while telling citizens to petition Congress, knowing full well that Congress is gridlocked by design, is not a defense of democratic self governance. It is a transfer of power from accountable institutions to an unelected body of nine. Sending abortion back to the states at the same time the Court has made it harder to challenge gerrymandered maps and voting restrictions is not an invitation to democratic participation. It is a trap.

The Chevron decision is the clearest example of this dynamic. For four decades, when a federal statute was ambiguous, courts deferred to the expert agency Congress had charged with implementing it. The Environmental Protection Agency interpreted environmental law. The Securities and Exchange Commission interpreted securities law. The reasoning was straightforward. Congress writes broad statutes because it cannot anticipate every situation. The agencies staffed with scientists, economists, and subject matter experts are better positioned than federal judges to fill in the details. By eliminating that deference, the Court did not return power to Congress. Congress still writes the same broad statutes. What changed is that federal judges, who are not experts in anything except law, now substitute their own judgment for the judgment of the people who actually understand the regulatory questions at hand. The practical effect is that regulated industries can challenge any rule they dislike in court, before judges who may have no background in the relevant science or economics, and win.

A conservative Court would restrain itself. It would respect precedent. It would be wary of concentrating too much power in any single institution, including its own. What we have instead is a Court that has systematically expanded its own authority while shrinking the authority of every other institution it touches. Federal agencies lose deference. States gain power over individual rights, but only the rights this particular majority is comfortable discarding. The channels for democratic correction, voting rights, campaign finance reform, redistricting, have been narrowed by the same Court that tells dissatisfied citizens to use democracy to fix what they do not like.

The Federalist Society provided the intellectual architecture and the personnel pipeline. Leonard Leo and his network of anonymous donors provided the money. Mitch McConnell provided the raw political will to break norms that had held for over a century. Donald Trump provided the popular constituency, millions of working class voters who had no stake in deregulation or corporate liability shields but who trusted him when he said he was fighting the establishment. And originalism, the theory that began as a legitimate call for judicial humility, provided the intellectual veneer for reaching outcomes the movement wanted all along, outcomes that lack majority support and could never be achieved through legislation. Dark money flows on both sides of American politics, through vehicles like the Arabella Advisors network on the left and the Koch and Leo networks on the right. But the scale, the coordination, and the singular focus on capturing an entire branch of government distinguish the conservative legal project from anything its counterparts have attempted.

The deeper problem is legitimacy. The Supreme Court has no army. It has no budget to speak of. Its authority rests entirely on the public’s belief that it is acting in good faith, interpreting the law rather than imposing a political program. When six justices appointed through a process shaped by one political faction, funded by anonymous donors, and installed through norm breaking maneuvers consistently deliver the outcomes that faction wants, that belief erodes. The institution that once commanded reverence from people like my family in Beaver County, people who did not follow legal theory but trusted that the system was fair, is losing the one thing it cannot function without.

The Warren Court was not perfect. Its critics accused it of overreach, of substituting its own moral convictions for the text of the Constitution, and some of those criticisms fed the very movement that would eventually capture the Court in the name of restraint. But the Warren Court was reaching toward something. It was trying to make the promise of equal protection real for people who had been excluded from it. The direction was outward, toward more liberty, more equality, more participation in the democratic project.

The current Court is reaching in the opposite direction. Inward, toward a narrower conception of who the Constitution protects and what the government can do on behalf of its citizens. It is consolidating power in the judiciary while dismantling the tools ordinary people use to hold their government accountable.

And the Court was only the beginning. While it cleared the legal obstacles, the Heritage Foundation and its network of conservative organizations were building something even more ambitious. They called it Project 2025, a 922 page blueprint for restructuring the entire executive branch of the federal government, written by hundreds of conservative operatives, nearly half of the collaborating organizations funded through Leonard Leo’s dark money network. The plan called for replacing tens of thousands of career civil servants with ideological loyalists, gutting regulatory agencies, dismantling environmental protections, and concentrating unprecedented power in the presidency. It was not a set of policy suggestions. It was an instruction manual for authoritarian governance, drafted in plain sight and published more than a year before the 2024 election.

Trump told the country he had never heard of it. “I know nothing about Project 2025,” he posted on Truth Social in July 2024. “I have no idea who is behind it.” This was a lie so transparent it barely qualifies as one. At least 140 people who worked in his first administration helped write the thing. Six of his former Cabinet secretaries contributed. His own running mate wrote the foreword to the Heritage Foundation president’s book. In April 2022, Trump had stood at a Heritage Foundation dinner and told the room they were “going to lay the groundwork and detail plans for exactly what our movement will do” when they returned to power. He knew. He always knew. But the voters who cheered for him at rallies in swing states did not know, and that was the point.

Within days of taking office in January 2025, the mask came off. A TIME analysis found that nearly two thirds of Trump’s initial executive orders mirrored or partially mirrored proposals from Project 2025. By the one year mark, the Center for Progressive Reform documented that the administration had initiated or completed 53 percent of the blueprint’s domestic policy agenda. The people who wrote it were appointed to implement it. Russell Vought, who authored the Project 2025 chapter on the Executive Office, was nominated to lead the Office of Management and Budget. Brendan Carr, who wrote the chapter on the Federal Communications Commission, was nominated to run it. The Heritage Foundation’s president, Kevin Roberts, declared that the organization’s role was to “institutionalize Trumpism.” This was not governance. It was an installation.

The people who voted for Trump because they were angry about their grocery bills, about fentanyl, about a country that seemed to have forgotten them, did not vote for any of this. They did not vote to strip the agencies that keep their water clean and their workplaces safe. They did not vote to replace career scientists and inspectors with political operatives whose only qualification was loyalty to one man. They did not vote to hand $1.6 billion in dark money influence over the structure of American democracy. They voted because a man who owned a gold plated penthouse in Manhattan looked into a camera and told them he understood their pain, and they believed him, because the resentment was real even if the messenger was not. He fed that resentment the way a con artist feeds trust. He gave it language, gave it targets, gave it the energy of a movement. And while the crowds roared, the operatives behind him quietly executed a plan to remake the government in ways that will harm the very people who put him there for a generation.

That is the transaction at the heart of this story. Not Trump and the donors. That deal was straightforward. The deeper transaction was between Trump and the tens of millions of Americans who traded their democratic leverage for the feeling of being heard. They were not heard. They were used. Their passion was converted into judicial appointments, regulatory rollbacks, and the installation of an authoritarian policy framework that no candidate could have won an election by honestly describing. The conservative movement did not persuade the public to support Project 2025. A poll by NBC in September 2024 found that just seven percent of registered Republicans viewed the plan positively. So they hid it behind a populist who promised to drain the swamp while filling it with ideologues who had been waiting decades for exactly this opportunity.

The donors saw it. The operatives planned it. The Federalist Society vetted the judges. The Heritage Foundation wrote the playbook. And Donald Trump sold it to people who never would have bought it if they had understood what they were purchasing.

When I was a kid and I heard others talk about the Supreme Court, they talked about it the way they talked about the church. It was a place where, when everything else failed, someone would do the right thing. That faith was not naive. For a period in American history, it was justified. What we have now is something different. The Court has been captured. The executive branch is being restructured by a blueprint the president pretended not to recognize. And the people whose resentment made it all possible are the ones who will pay the highest price for what comes next. Pretending otherwise is not respect for the institution. It is complicity in its transformation into something the founders would not recognize and should not have had to imagine.

Here is the story that got us started on the topic…

“But why” was what I was after in this conversation. What the money is actually buying. And who, specifically, built the machine that makes it possible.

I want to lay this out, because the answer to that "why" is the most important part…

I want to be clear about something before I go any further. Lobbying itself is not the problem. Congress writes laws that affect every industry in the country, and legislators cannot possibly be experts in all of them. Lobbying, at its core, is the process of educating lawmakers so they understand the consequences of what they are voting on. That is not just legitimate. It is necessary. But what is happening in the United States right now is not education. It is purchase. The system has been perverted into something that would be unrecognizable to anyone who designed it, a pay-to-play machine where access is sold, legislation is auctioned, and the line between informing a legislator and owning one has been erased entirely. That perversion is a cancer on the country, and it is eating American democracy from the inside.

Start with the raw numbers. In 2025, U.S. federal lobbying spending hit a record $5.08 billion, a 14 percent jump in a single year, the largest increase since quarterly disclosures began in 2008. The year before, the total was $4.4 billion. Now compare that to the European Union, which has a comparable population and a comparable number of registered lobbyists. The EU's Transparency Register shows estimated lobbying expenditures of roughly $1.5 to $2.2 billion per year. The U.S. routinely spends two to three times more in raw terms. But it is not the raw gap that matters. It is what the money buys.

A University of Kansas study examined what happened when corporations lobbied for a one-time tax holiday in 2004 that let them bring overseas profits home at a drastically reduced rate. The researchers found that for every dollar spent on lobbying, those corporations received $220 in tax savings. That is a 22,000 percent return on investment. Ninety-three firms spent a combined $282.7 million lobbying for the provision and collectively saved $62.5 billion. Eli Lilly alone spent $8.52 million lobbying and received more than $2 billion in tax savings. A separate Sunlight Foundation analysis found that, on average, the most politically active corporations in America received $760 from the government for every dollar they spent on influencing politics.

That is the answer my friends and I kept circling back to. That is the "why" behind everything else in this article. The money flooding American politics is not ideological passion. It is not civic participation. It is an investment, and it is the highest returning investment in the American economy. Corporations spend billions on lobbying and campaign contributions because the exposed regulatory and tax machinery of the U.S. government pays them back in trillions. Every loophole, every favorable regulation, every killed enforcement action, every watered-down environmental rule has a price tag attached to it, and the corporations paying for access have done the math. They know exactly what they are buying.

The European Union, for all its imperfections, understood this. Most EU member states treat political money as a vector for corruption and regulate it accordingly. France bans corporate donations to political parties outright. So do Lithuania, Romania, Slovakia, and thirteen other EU member states. Germany caps corporate donations at 500 euros. Italy caps them at 500 euros. The Netherlands caps them at 1,000 euros. Most EU countries ban anonymous donations above small thresholds. Most impose spending limits on campaigns. The United States and Finland are the only OECD countries with no spending limits at all. Sweden and Norway rely heavily on public subsidies and severely restrict large private donations specifically to insulate campaigns from corporate interests. France makes parties largely reliant on public funding.

The design principle behind all of this is simple. If you let corporations buy legislators, they will. If you let the return on political investment exceed the return on productive investment, corporations will rationally spend more money buying politicians than building products. European democracies looked at that math and said, we need to break that equation. The United States looked at the same math and said, free speech.

The single most destructive act that made this possible was the Supreme Court's 2010 decision in Citizens United v. Federal Election Commission. The case was brought by a conservative nonprofit organization called Citizens United, founded in 1988 by Republican political consultant Floyd Brown and run since 2000 by David Bossie, who later became Donald Trump's deputy campaign manager. The organization had accepted funding from the Koch brothers, Charles and David Koch, the billionaire industrialists who had spent decades and hundreds of millions of dollars funding a network of conservative advocacy groups, think tanks, and legal organizations designed to dismantle campaign finance regulation. Koch-connected organizations including the Cato Institute, the Institute for Justice, and the Center for Competitive Politics filed amicus briefs supporting the case. The Kochs did not invent Citizens United, but they had spent thirty years building the legal and intellectual infrastructure that made it possible.

And they had a very specific reason for doing it. Koch Industries is the second largest privately held company in the United States. It is an oil, gas, chemical, and manufacturing conglomerate with operations that generate enormous pollution and face enormous regulatory exposure. In 2000, the EPA imposed the largest civil fine ever levied on a company under any federal environmental law, $30 million, after Koch was found responsible for more than 300 oil spills in six states that leaked some three million gallons of crude into ponds, lakes, rivers, and streams. In 2001, Koch pleaded guilty to Clean Air Act violations at a Texas refinery for covering up illegal benzene emissions and paid another $20 million. A subsidiary paid $500 million to fix environmental violations at facilities in seven states. Koch Industries has been hit with 150 separate state environmental penalties across every state where it operates. The company was also named one of the top ten air polluters in the United States. When you are a company that pollutes that much, environmental regulation is an existential cost. Climate policy is an existential threat. Campaign finance restrictions are the lock on the door that keeps you from buying the legislators who write those rules. The Kochs did not fund thirty years of anti regulation legal infrastructure out of philosophical conviction about free speech. Maybe they did it because they needed the regulations that cost them billions to go away, and the cheapest way to make that happen was to buy the political system that writes them. Just maybe, dismantling campaign finance law was not the goal, but it was the tool.

Is the investment paying off:

• January 20, 2025. Trump signed "Unleashing American Energy," revoking Biden-era climate and environmental executive orders and ordering a review of all environmental regulations.

• January 31, 2025. Trump signed an executive order requiring ten existing regulations be eliminated for every new one issued.

• February 2025. The EPA rescinded its NEPA implementing regulations, gutting the government's primary tool for requiring environmental impact reviews.

• March 2025. The EPA announced plans to roll back 31 environmental rules on clean air, clean water, and climate change. The agency also began reconsidering wastewater regulations for oil and gas industries.

• March 2025. Americans for Prosperity, the Koch network's flagship political arm, launched a $20 million campaign to push Trump's tax cut agenda through Congress.

• June 2024 (still rippling). The Supreme Court overturned the Chevron doctrine, a Koch network priority for decades, stripping federal agencies of their authority to interpret ambiguous environmental statutes. Koch-connected groups filed amicus briefs. Koch-backed Americans for Prosperity had spent money supporting the confirmation of the three Trump-era justices who joined the majority.

• July 4, 2025. Trump signed the One Big Beautiful Bill Act, cutting $5 trillion in taxes over ten years. Corporate tax provisions include permanent 100 percent bonus depreciation, expanded R&D expensing, and a reduced 15 percent corporate rate for domestic manufacturing. Koch Industries, as a massive manufacturer and refiner, is a direct beneficiary.

• Fall 2025. The EPA moved to repeal the 2009 Endangerment Finding, the legal foundation for all federal greenhouse gas regulation under the Clean Air Act. EPA Administrator Zeldin called it "driving a dagger straight into the heart of the climate change religion."

• December 2025. The EPA proposed stripping Clean Water Act protections from up to 70 million acres of wetlands and potentially millions of miles of streams.

The five justices who voted in the majority were Anthony Kennedy, who wrote the opinion, John Roberts, Antonin Scalia, Clarence Thomas, and Samuel Alito. Together they ruled 5-4 that political spending is a form of protected free speech, and that restricting corporate expenditures on elections therefore violates the First Amendment. Justice John Paul Stevens, joined by Ruth Bader Ginsburg, Stephen Breyer, and Sonia Sotomayor, called the decision "profoundly misguided" and warned it would undermine the integrity of elected institutions across the nation.

The results were immediate and staggering. Outside group political spending exploded roughly 28-fold from 2008 to 2024, from $144 million to $4.21 billion. Dark money from groups that do not disclose their donors reached $1.9 billion in the 2024 election cycle alone, according to the Brennan Center for Justice, nearly double the previous record and up from less than $5 million in 2006. Since the decision, dark money groups have spent at least $4.3 billion on federal elections. The five justices who voted for this did not just open a door. They blew out the wall.

And then look at what walked through. In the 2024 presidential election, Elon Musk, the world's richest person, spent more than $250 million to help elect Donald Trump. To understand why this matters to someone in the EU, where this kind of thing is literally illegal, let me explain exactly what Musk did. He created a super PAC called America PAC, which is a type of political spending vehicle that can raise and spend unlimited money thanks to Citizens United. Musk personally provided 91 percent of the group's funding. A March 2024 FEC ruling had determined that door-to-door canvassing falls outside the ban on coordination between super PACs and campaigns. So Musk's PAC was legally permitted to coordinate its ground operations directly with the Trump campaign. The campaign essentially outsourced its voter turnout operation in all seven major swing states to a billionaire's private organization. America PAC ran the door-knocking, the text message campaigns, the digital ads, the direct mail. Musk also ran a daily $1 million cash giveaway to registered voters who signed a conservative petition, an operation the Justice Department warned might be illegal but which a judge allowed to continue. He also owned the social media platform X, which amplified his pro-Trump posts to every subscriber on the platform. In France or Germany, every single one of these activities would have been prohibited by law.

But Musk is not the disease. He is the symptom. The disease is a system where that kind of investment is rational. Musk now has the ear of the president, his companies hold billions in government contracts, and the regulatory environment he operates in is shaped by the administration he helped install. The return on his $250 million political investment will be measured in the hundreds of billions.

The next piece of the machine is the revolving door. In the U.S., moving between Congress, federal agencies, and lobbying firms is a recognized career path. More than 5,000 former government officials are currently registered as lobbyists. Research shows that each additional unit of Congressional staff connections adds $155,000 to $360,000 in annual lobbying revenue for revolving door lobbyists. The network is literally monetized. Former members of Congress face only a one-year cooling-off period before they can lobby Congress directly. Senators Mitch McConnell and Chuck Schumer have each had 65 former staffers pass through the revolving door individually. That is not an accident. That is a pipeline.

And the firms on the receiving end are not subtle about what they are purchasing. Ballard Partners, a lobbying firm founded by major Trump fundraiser Brian Ballard, hauled in more than $88 million in 2025, a 350 percent jump from the previous year. The firm previously employed Trump's White House Chief of Staff Susie Wiles and his Attorney General Pam Bondi. When your former chief of staff and your former attorney general both worked at the same lobbying shop, the revolving door is not a metaphor. It is a business model.

France's ethics authority, the HATVP, actively reviews and blocks public-to-private moves it deems conflicts of interest. Canada's Lobbying Commissioner has powers to investigate and penalize revolving door violations. The EU can revoke lobbying access credentials. These systems are imperfect, but they are functional.

And here is the part that sealed the whole thing shut. The United States has a federal agency tasked with enforcing campaign finance laws called the Federal Election Commission. It is structurally designed to fail. The FEC has six commissioners. It requires four votes for any enforcement action. By law, no more than three can be from the same party. The result is that the commission deadlocks on enforcement more often than not, frequently declining to even investigate alleged violations. Former commissioners have openly accused colleagues of refusing to enforce the law. A March 2026 filing before the D.C. Circuit noted that a partisan minority of commissioners have used procedural loopholes to kill enforcement matters entirely, creating what critics call a superpower to block accountability with zero judicial oversight. RUSI, the Royal United Services Institute in London, recently described the FEC as dysfunctional and ineffective.

So here is the full architecture, and the reason I said the "why" is the most important part. It is a closed loop with no exit. Corporations fund elections through PACs and super PACs to elect favorable legislators. Those legislators write the tax code, the regulations, and the contracts that pay the corporations back at 22,000 percent. The legislators' staffers leave government and walk across the street to K Street, where they sell the access and relationships they built on the public payroll. The lobbying firms use those connections to secure more favorable treatment for their corporate clients. And if anyone gets caught violating the rules along the way, the FEC is built to deadlock and do nothing about it. Other democracies break at least one link in that chain. They ban or cap corporate donations. They impose longer cooling-off periods. They publicly fund parties. They build enforcement agencies that actually work. The United States is almost unique in allowing every link to operate simultaneously, at massive scale, with a Supreme Court that declared the arrangement constitutionally protected and a regulator designed to look the other way.

I have been in Portugal for a few months now, and one of the things that strikes me most about European civic life is the assumption that democracy requires active protection from concentrated money. It is not a radical idea here. It is a design principle. The Portuguese, the French, the Germans, the Scandinavians, they all built guardrails because they understood what happens when you let the return on political investment exceed the return on building something real. They understood it from their own histories of authoritarian capture and democratic collapse.

Americans were not too naive to build those guardrails. We had them. We had a century of campaign finance law, from the Tillman Act of 1907 through the Bipartisan Campaign Reform Act of 2002. And then five justices tore them down, a network of billionaire industrialists funded the legal machinery that made it possible, a political class that profits from the wreckage refused to rebuild, and the agency charged with policing all of it was built with a kill switch that its own commissioners use regularly.

This is not a story about American decline as some mysterious force. It is a story about an investment thesis. Corporations and billionaires realized that the highest return available in the American economy was not in innovation, not in products, not in services, but in purchasing the regulatory and tax machinery of the government itself. And when other democracies looked at that math and built walls against it, the United States tore its walls down and called it freedom.

And now the United States is doing the same thing to AI policy… giddy up…

Decades later, when millions of Americans looked at Washington and decided the whole thing was a swamp that needed draining, I understood the impulse. These were not ignorant people being manipulated. They were people who had been watching a rigged game for thirty or forty years and had finally been given a word for what they were seeing.

But here is what I have come to believe, and what I think this country needs to sit with before it can find its way to anything better. The word was wrong. Not the anger. The anger was earned. The diagnosis was wrong. And a wrong diagnosis, pursued with righteous conviction, does not cure the disease. It feeds it.

The question nobody asked honestly, the one that would have changed everything if they had, is this. Is what we have been living under actually capitalism? Or is it something else entirely, something older and more familiar, wearing capitalism’s clothes?

Capitalism in its theoretical form requires a few things to function. Open competition. The ability of new players to challenge established ones. Price signals that reflect genuine supply and demand. Creative destruction, where bad ideas and inefficient companies fail so that better ones can replace them. And most importantly, the assumption that outcomes are determined primarily by what you offer in the marketplace rather than by who you know in the back room.

Hold that framework in your mind and then look at what actually exists.

When a pharmaceutical company spends more on lobbying than on research and then gets Congress to prohibit Medicare from negotiating drug prices, that is not the capitalism we claim to practice. When a bank takes catastrophic risks, fails, and gets rescued with public money while the homeowners it defrauded lose everything, that is not the capitalism we teach in textbooks. Capitalism demands that failure carries consequences. What happened in 2008 was something else. It was a system ensuring that consequences flowed downward while protection flowed upward. When a handful of companies acquire every competitor in a sector until three or four players control an entire market, and the antitrust apparatus meant to prevent this has been deliberately starved and ideologically captured, that is not competition. That is consolidation protected by proximity to power.

Some economists would call this crony capitalism or political capitalism and argue that the merger of capital accumulation with political access is not a corruption of the system but an endogenous feature of it, something markets tend to produce when left to compound long enough. They may be right. But the distinction matters less than the recognition. Whether you believe market capitalism inevitably generates an influence economy or that the influence economy is a parasite that attached itself to an otherwise functional host, the diagnosis leads to the same place. What we are living under is not the system Americans think they are defending when they defend free enterprise. It is something else. And until we name it clearly, we cannot begin to fix it.

What we actually have, and what we have had for decades, is capitalism fused with an influence economy so deeply that the two have become indistinguishable. And understanding how that influence economy works is the key to understanding why draining the swamp was always going to fail, and why what replaced it was always going to look like what came before.

An influence economy runs on relationships, not products. Its currency is access, not innovation. And it operates at every level of American life, not just in Washington.

In DC, a retired four-star general joins a defense contractor’s board. He does not design weapons systems. His value is that he can pick up the phone and reach a current Pentagon official who will take the call. A Senate staffer helps a lobbyist get favorable language into a committee report. Two years later, that staffer needs a job and the lobbyist makes an introduction at a firm. Nobody explicitly traded anything. But both understood the ledger. A technology company hosts “educational briefings” for congressional staff on artificial intelligence policy. Free lunch, credentialed speakers, nice venue. The company is not buying votes. It is becoming the default source of expertise on a topic so that when legislation gets drafted, staffers reach for the frameworks the company already handed them.

But this is not just Washington. The Ivy League runs on the same currency. A kid who goes to Harvard is not just learning economics. That kid is sitting next to the children of senators, hedge fund managers, and media executives. The friendships formed at nineteen become the professional relationships that open doors at thirty-five. For many families, that network is the product they are purchasing. Legacy admissions are not about tradition. They are about maintaining the network’s continuity across generations.

Corporate America operates identically. A small number of people sit on multiple boards, and they hold those seats not because they possess unique expertise in each company’s industry but because they are trusted members of the club. The golf outings, the Davos dinners, the Sun Valley conferences are not recreation. They are where the informal consensus gets built that later shows up as a merger announcement or a coordinated lobbying push or a quiet agreement about executive compensation norms. CEO pay keeps climbing despite shareholder complaints because the boards that set compensation are populated by other executives who have every incentive to keep the benchmarks high.

This is the actual operating system of American economic life. Not supply and demand. Not creative destruction. Not meritocratic competition. Relationships. Access. Position. The slow accumulation of favors and obligations that compound over time into durable structural advantage.

And here is the part that should unsettle everyone, because it means the problem is deeper than anyone in politics wants to admit. This pattern is not a uniquely American corruption. It is not even a modern phenomenon. The raw materials for building influence networks are pre-installed in every human brain, and every complex society in recorded history has assembled some version of them.

Trust is biologically expensive. It takes time, repeated interaction, and shared vulnerability to build. Because it is costly, it becomes scarce, and scarce things become valuable. Once you have built a trust network, you protect it. You limit entry. You favor the people inside it. Not because you are malicious but because your nervous system treats the extension of trust to strangers as a risk, and it is right to do so.

Reciprocity runs deeper than reason. Someone does something for you and you feel an almost physical need to return it. This is the engine of every influence economy that has ever existed, from the Roman patronage system to K Street. It is not rational calculation. It is an emotional reflex that evolved because cooperative groups survived and non-cooperative ones did not.

Status seeking is the accelerant. Humans monitor their position in hierarchies with obsessive precision, and they pursue advantage because for most of evolutionary history, higher status meant survival. The accumulation of influence is just status competition expressed through social architecture rather than physical dominance.

These impulses are universal and enduring. But the institutions they produce are not identical. The Medici patronage system, the British class hierarchy, the Chinese tradition of guanxi, the American revolving door between government and industry, these are all members of the same family, shaped by the same underlying drives, but expressed through vastly different institutional forms depending on the legal frameworks, cultural norms, and economic structures of each society. The drives are inevitable. The specific shape they take is not. And that distinction is everything, because it means institutional design matters. It means the structures a society builds around these impulses determine whether they produce a functional economy or a closed loop of self-dealing.

The societies that have functioned best are not the ones that eliminated these dynamics. That is impossible. They are the ones that built countervailing structures to contain them. Independent courts. Free press. Public education that creates alternative pathways. Antitrust enforcement that prevents economic concentration from becoming political concentration. Campaign finance systems that limit the conversion of wealth into access. These structures do not drain the swamp. They build levees around it. And they require constant, unglamorous maintenance.

The United States had most of these structures at various points in its history. And then, over the last fifty years, it systematically weakened every single one. The press was consolidated. Public education was starved. Antitrust enforcement was ideologically neutered. Campaign finance was deregulated. With each weakening, the influence networks grew stronger, the public grew more frustrated, and the demand for a dramatic solution grew louder.

Enter “drain the swamp.”

The slogan was genius as politics and catastrophic as diagnosis. It validated the legitimate anger of millions of people while channeling that anger toward a solution that could never work.

What the slogan did, with devastating efficiency, was reframe a systemic problem as a personnel problem. It told people that the issue was not the structure of the influence economy but the specific people operating it. The bureaucrats. The career politicians. The deep state. Remove them, install new people, and the problem would be solved. This is the move that guarantees failure, and not just in this instance. Any movement that promises transformation purely through better people, without reforming the institutional incentives, the campaign finance rules, the civil service protections, the transparency requirements that shape how those people behave once in office, is structurally destined to reproduce the same influence dynamics it claims to oppose. The faces change. The architecture does not.

The coalition that believed this was enormous and sincere. It included the steelworker who watched the mill close while Washington debated things that had nothing to do with his life. The small business owner who could not understand why regulations seemed designed for companies with compliance departments she could never afford. The veteran who came home to a bureaucracy that could not process a disability claim. The young person who did everything right, took the degree, accepted the debt, and still could not find a foothold in an economy that rewarded connections over competence.

These people were not wrong about the problem. They were given a theory of the problem that was incomplete in ways that guaranteed the solution would fail. Because the swamp is not a collection of bad actors who can be fired and replaced. The swamp is what emerges when trust is scarce and complexity is high and resources flow toward whoever stands closest to the machinery of decision. You can replace every person in the building. The dynamics reassert themselves within a single election cycle because the conditions that produce them have not been touched.

So what would actually draining the swamp have required?

If you took the grievance seriously, if you genuinely wanted to break the influence economy’s grip on American governance, there was a real agenda available. It would have started with campaign finance reform that severed the direct pipeline between private wealth and political access. It would have included enforceable lobbying restrictions with real teeth, not the current system where a lobbyist can re-register as a “strategic consultant” and continue the same work without disclosure. It would have meant strengthening civil service protections so that the permanent government workforce operated on competence and institutional knowledge rather than political loyalty. It would have required transparency mandates that forced every meeting between industry and regulators into the public record. It would have demanded antitrust enforcement vigorous enough to prevent the kind of market concentration that converts economic dominance into political leverage. And it would have invested in the public institutions, the schools, the courts, the press, the regulatory agencies, that serve as the countervailing structures against network capture.

None of this is exotic or untested. These are the tools that every functioning democracy uses to manage the influence dynamics we have been discussing. They are boring. They are incremental. They do not fit on a hat or fill an arena. But they work, in the specific sense that they keep the levees intact.

Now look at what was actually done.

Campaign finance was not reformed. It was left untouched while the wealthiest individual in human history was given direct operational authority inside the federal government. Lobbying restrictions were not strengthened. The revolving door was not closed. It was ripped off its hinges and replaced with a direct portal between Silicon Valley and the executive branch. Civil service protections were not reinforced. They were systematically dismantled. Hundreds of thousands of career professionals, the people who understood how agencies functioned and who provided institutional continuity across administrations, were fired, placed on indefinite leave, or pressured into resignation. Transparency was not expanded. Communications moved to Signal. Decision-making authority shifted to unelected individuals operating outside normal chains of accountability. Antitrust enforcement was not strengthened. The same concentration of economic power that created the problem was invited to direct the solution. And the public institutions that serve as checks on network capture, the regulatory agencies, the inspectors general, the independent research bodies, the foreign aid apparatus, were not reformed. They were gutted.

Every single item on the structural reform agenda was not merely ignored. It was inverted. The actions taken did not weaken the influence economy. They removed the last remaining barriers to its acceleration. They did not drain the swamp. They paved it, built on it, and handed the deed to a new set of landlords.

There is a word for what happens when you promise one thing and deliver its opposite while using the language of the original promise to prevent anyone from pointing out the switch. It is called a bait and switch. And I use that term not as rhetoric but as a structural description. The bait was reform. The switch was capture. And the people who swallowed the bait are the same people who are now paying the cost of the switch.

And the receipts are now public.

In early 2026, court depositions and discovery documents became public in several federal lawsuits that revealed how the Department of Government Efficiency was actually constructed and how it operated inside the agencies it was sent to transform. The picture that emerges, and I want to be precise here because these are allegations in active litigation, not final legal findings, is one that should be deeply recognizable to anyone who has followed this argument. Every single feature of the old influence economy is present. It is just wearing different clothes.

A young man who graduated from the University of Virginia in 2020 and worked in private equity got into DOGE through a friend whose father happened to be the chief financial officer at one of Elon Musk’s companies. Another recruit was introduced by a venture capital fund manager who served as an “informal recruiter.” That recruit had no government experience. He had never taken a government-related class. His qualification was proximity to the right network.

The recruiting pipeline ran through Musk’s personal orbit and Peter Thiel’s circle. Marc Andreessen, the billionaire venture capitalist, worked as a self-described “unpaid intern” screening candidates. Internal communications ran on Signal, a choice that a federal judge would later note appeared to violate the Federal Records Act. The official head of DOGE, a career government employee named Amy Gleason, apparently never led a single meeting. Real authority flowed through Steve Davis, a Musk loyalist, in a shadow structure that one official described under oath as feeling “more like a club” than a government operation.

A club. The word should stop you cold.

Not an agency with transparent hiring practices. Not an institution with accountability mechanisms and public oversight. A club. Membership determined by who you knew. Authority flowing through personal loyalty rather than legal mandate. Access granted on the basis of proximity to wealth rather than relevant expertise or public mandate.

And the consequences of club governance were not abstract. At the National Endowment for the Humanities, according to court filings, these DOGE associates fed grant descriptions into ChatGPT and asked it to determine which ones related to diversity, equity, and inclusion. The AI flagged 1,057 out of 1,163 grants it reviewed. The results were entered into a spreadsheet that became the basis for canceling more than $100 million in previously awarded funding. Among the grants terminated were a documentary about Jewish women’s slave labor during the Holocaust, a project to digitize photographs of Appalachian residents, efforts to preserve endangered Native American languages, and a $349,000 grant to replace a museum’s aging HVAC system. The NEH’s own acting chair warned in an email that many of the projects on the chopping block had no justification for cancellation. According to the depositions, his objections were overridden. The DOGE associates, neither of whom had consulted scholars or the NEH’s peer review system, made the final calls.

At USAID, the pattern scaled to global consequence. DOGE’s involvement in dismantling the sixty-year-old foreign aid agency contributed to the termination of what the State Department acknowledged was at least 83 percent of its program contracts and the departure of roughly 10,000 staff. A federal judge found that the dismantling likely violated the Constitution. A public health researcher at Boston University estimated that the resulting cuts to foreign aid programs contributed to more than 700,000 excess deaths by January 2026, a figure that continued to climb. A separate analysis found that the supposed savings were illusory, that DOGE’s cuts may have actually increased costs to taxpayers by approximately $135 billion when accounting for the expense of firing, rehiring, and placing workers on paid leave. Government spending, according to the Cato Institute, actually rose under DOGE.

The favor bank. The social positioning. The network as currency. The conversion of private wealth into public authority without democratic consent. Every mechanism of the influence economy we have traced through Washington, through the Ivy League, through corporate boardrooms, and all the way back to the evolutionary roots of coalition-building, was operating at full speed inside the very institution that promised to end it all. And the downstream consequences were not theoretical. They were measured in shuttered research projects, dismantled agencies, and according to one estimate, hundreds of thousands of lives.

This is where the story becomes genuinely tragic, because the people who believed the promise are the ones paying the price.

The steelworker in Beaver County did not vote for a system where twenty-somethings recruited through Silicon Valley dinner parties would decide which government programs survive. The small business owner did not vote for the world’s richest man to embed his personal associates inside federal agencies with access to sensitive data systems. The veteran did not vote for a parallel governance structure accountable to a billionaire’s social circle rather than to the public.

But pointing this out gets you accused of defending the old swamp. The framing has been constructed so that any critique of the new network reads as nostalgia for the previous one. You are either for the revolution or you are part of the problem. This false binary makes honest conversation almost impossible, and that is not an accident. It is the condition under which influence networks have always thrived. They need the silence. They need the tribal loyalty. They need people arguing about which team should run the system rather than examining the system itself.

And the disillusionment that follows, the slow recognition that the new arrangement feels indistinguishable from the old one, does not produce clarity. It produces cynicism. People stop believing reform is possible. They disengage. And cynicism is the one state under which influence networks become truly untouchable, because the public has stopped watching and the insiders no longer need to pretend.

I write this from Cascais, Portugal, where I have lived for a few months. I mention this not to claim some superior vantage point but because distance clarifies things that proximity obscures. Portugal has its own patronage networks, its own insider dynamics, its own legacy of institutional capture. But it also has something the United States has been losing for a long time. It has civic infrastructure that has not been fully hollowed out. Its national health service, while strained, still provides universal coverage funded through general taxation rather than through a labyrinth of private intermediaries. Its public transit system connects cities and regions at a scale that most American communities lost decades ago. Its regulatory environment, while far from perfect, has not been entirely colonized by the industries it oversees. These outcomes are not accidental. They are the product of specific institutional choices, of a society that decided the countervailing structures were worth maintaining even when they were expensive and unglamorous and politically unrewarding.

The United States faces a version of this choice right now, and the stakes have never been higher. Because the next great influence network is already being built, and it is being built around artificial intelligence. A very small number of companies and individuals are constructing the technological infrastructure that will shape economic outcomes for the next generation. The influence networks are already forming around who controls these systems, who gets access, and whose values get encoded into the technology itself. The lobbying, the revolving doors, the strategic positioning, all of the dynamics we have traced through this entire essay are already in motion around AI governance, and they are moving faster than any countervailing structure can currently keep pace with.

The question is not whether the influence economy will colonize this new territory. It will. It already has. The question is whether this time, finally, enough people understand the pattern clearly enough to build the levees before the flood. And the fear, the one I cannot shake and that the evidence does nothing to quiet, is that we will not act at the scale or speed required.

The people who wanted to drain the swamp deserved better than what they got. They deserved an honest accounting of what the swamp actually is. Not a group of corrupt politicians. Not a deep state conspiracy. Not a partisan enemy that can be defeated in a single election. The swamp is what happens to every complex society when the biological impulse to build influence networks is allowed to compound without structural constraint. It is a species, not a scandal. It adapts. It evolves. It colonizes whatever institution you build to contain the last version of it.

You cannot drain it. You can only build the structures that hold it in check, and then you must maintain those structures forever, because the moment you stop, the water rises again.

That is not a slogan anyone will ever put on a hat. But it is the truth. And I think this country has had enough of the alternative.

The Pentagon says it would never use AI for mass surveillance. Anthropic put that promise in the contract. And the Pentagon designated them a national security threat for doing it.

But why would a government go to war with a company over a contract clause it claims it already keeps? If they won’t tolerate that limit in a contract they can renegotiate, they certainly won’t tolerate it trained into the AI’s value system (RLHF, Constitutional AI, and other fine tuning methods), where no customer can override it at the point of use.

Does it seem like this administration wants control of that value system?

To understand why that question matters, you first have to understand what a value system inside an AI actually is, how it gets there, and what is already being done to change it.

Every major AI lab is currently hard coding a value system into its models. Not a preference. Not a tendency. A value system. And they are each doing it differently, through processes that are almost entirely invisible to the public, to regulators, and to the billions of people who will soon be relying on these systems to help them think.

At OpenAI, the process works like this. After a model like GPT is pretrained on massive amounts of internet text, it goes through a second phase called Reinforcement Learning from Human Feedback. OpenAI hires teams of human contractors, roughly 40 in the original InstructGPT program, who are trained to evaluate the model’s outputs. They compare pairs of responses and choose which one is better. Those preferences are used to build a reward model, essentially a scoring system that reflects the contractors’ judgments about what counts as helpful, accurate, and appropriate. The language model is then fine tuned to maximize that score. In practice, small teams of people, largely based in San Francisco and hired through a screening process, play an outsized role in defining what “good” means for a system that now serves hundreds of millions of users. OpenAI has since published a Model Spec that describes its intended behavioral guidelines, but the document explicitly notes that human labelers and researchers use it to generate the RLHF training data that shapes the model’s actual conduct.

Anthropic takes a different approach. Rather than relying primarily on human raters, Anthropic developed a method called Constitutional AI. The company writes a set of principles, a “constitution,” and the AI model is trained to critique and revise its own outputs against those principles. The model asks itself questions like “Is this response encouraging violence?” and “Is this answer truthful?” and iterates until it satisfies the rules. Then a second AI model evaluates which outputs best comply with the constitution, and those preferences are used to further train the system. The human judgment is not eliminated. It is front loaded into the writing of the constitution itself. In January 2026, Anthropic published a new, comprehensive version of Claude’s constitution under a Creative Commons public domain license, making it the most transparent governance framework in the industry. But as a Lawfare analysis noted, the constitution “is unilaterally authored by designers, not by the users and individuals whom the AI’s actions may affect.” A small team of researchers decides what principles the AI should follow, and the machine then enforces those principles at scale without continuous human oversight.

And then there is xAI, Elon Musk’s company. Grok has been marketed from the beginning as “unfiltered,” a deliberate contrast to what Musk characterizes as “woke” AI. In practice, this has meant dramatically fewer safety guardrails. In one independent security test, Grok 4 scored near zero on several widely used safety benchmarks when tested without a system prompt. In documented red team exercises, researchers were able to elicit assistance with biological weapons related scenarios and obtain other harmful content with relatively few refusals. In one widely reported incident, Grok generated posts on X that included language described by critics and watchdog groups as antisemitic before the company intervened. By late 2025, CNN reported that users were able to use Grok’s image tools to create sexually suggestive deepfaked images, including images that appeared to depict minors, before xAI restricted the feature under global backlash. Meanwhile, Grok 4.1’s sycophancy rate in one benchmark was roughly three times that of earlier versions, meaning it was more likely to mirror user views even when those views were incorrect. The pattern reflects a foundational design choice about what values to encode and, more importantly, which ones to leave out.

Three companies. Three fundamentally different approaches to encoding moral judgment into systems that will increasingly shape how people understand the world. None of them were designed through any public process. In the United States, there is still no comprehensive, binding oversight regime for these systems. And as of January 2025, the most visible federal initiative, the Biden administration’s Executive Order on AI, was rescinded by President Trump on his first day in office.

Why does a hidden value system inside an AI matter?

Because these models are rapidly becoming the operating system for human decision making. When a bank evaluates a small business loan application, when a doctor works through a differential diagnosis for a rare disease, when a voter researches a candidate’s record, increasingly they are seeing the world through a filter designed by a small number of engineers at a handful of companies. The AI does not just retrieve information the way a search engine does. It synthesizes, prioritizes, frames, and in many cases recommends. It shapes what you see and how you see it. That is not a search function. That is an editorial function, operating at a scale no newspaper, television network, or social media algorithm has ever achieved.

Why does it matter that AI is becoming the operating system for human decisions?

Because unlike a calculator, which is mathematically certain and indifferent to its user, an AI is a judgment engine. A calculator does not care who punches in the numbers. It returns the same answer for a hedge fund manager and a high school sophomore. But an AI makes calls. It does not merely present data. It decides whether a protest constitutes legal dissent or public disorder. It determines whether a job applicant’s resume reflects potential or risk. It assesses whether a patient’s symptoms suggest a routine condition or something the doctor should worry about. Those judgments are not objective truths delivered from some computational oracle. They are the pre programmed echoes of whoever trained the model. At OpenAI, they reflect the preferences of a few dozen contractors selected through a screening test. At Anthropic, they reflect the principles chosen by a small research team. At xAI, they reflect a deliberate corporate decision to minimize guardrails in the name of being “unfiltered.” In every case, the judgments are someone’s. And in every case, that someone was not elected, appointed, or publicly accountable.

Why does it matter that AI makes judgment calls instead of calculations?

Because those judgments scale instantly to billions of people. A biased moral philosophy embedded in a single model can automate mass surveillance or economic exclusion at the push of a button. If a model’s underlying philosophy privileges state stability over individual privacy, a government can monitor and suppress millions of citizens without ever hiring a single human analyst. If a model’s training data reflects the assumption that certain neighborhoods are higher risk, every loan application from those zip codes gets quietly downgraded before a human being ever looks at it. This is not hypothetical. We have already seen how algorithmic bias in far simpler systems produced discriminatory outcomes in criminal sentencing, hiring, and credit. Now imagine that same dynamic operating through a system that is orders of magnitude more powerful, more deeply embedded in daily life, and whose internal value system was designed in private by people who face no consequences when it goes wrong.

Why does it matter that biased AI judgments scale to billions of people?

Because the people holding the kill switch over these value systems have no democratic mandate, and the government that is supposed to check their power is not just dismantling oversight. It is actively working to replace the labs’ value systems with its own.

The CEOs of OpenAI, Anthropic, Google DeepMind, and xAI are unelected architects of what is becoming global cognitive infrastructure. They did not run for office. They were not confirmed by any legislative body. No public deliberation shaped the principles they are encoding. They are making civilizational choices in private, under competitive pressure, with fiduciary obligations to investors, not citizens.

The primary check on this power is supposed to be government oversight. But the practical effect of the Trump administration’s approach has been to weaken public accountability and prioritize rapid deployment over safety. On his first day in office, President Trump rescinded the Biden administration’s AI executive order, the most comprehensive federal AI governance framework ever issued, which had required safety reporting for powerful models, mandated red teaming for high risk systems, and directed agencies to develop best practices for AI safety. Three days later, Trump signed a replacement order titled “Removing Barriers to American Leadership in Artificial Intelligence,” whose language repeatedly frames safety requirements and “engineered social agendas” as obstacles to innovation. The order contains no specific safety directives. It appoints David Sacks, a venture capitalist, as AI and Crypto Czar.

Then came the fight with Anthropic. To understand what it reveals, you need to see that there are three distinct layers of constraint on how an AI can be used, and the administration is now operating on all three of them.

The first layer is law and policy. Federal statutes, executive orders, and Pentagon directives that restrict what the military can do. These are real constraints, but they are controlled entirely by the government. A future Congress can rewrite a statute. A new president can rescind an executive order, as Trump did with Biden’s AI safety order on his first day in office. A Defense Secretary can revise a policy memo. The government sets these rules, and the government can change them.

The second layer is contract language. The terms of sale between an AI company and its customer. This is what the Anthropic dispute was explicitly about. In January 2026, Defense Secretary Pete Hegseth issued an AI strategy memorandum requiring that all Department of Defense AI contracts adopt standard “any lawful use” language. Anthropic had two contractual red lines, prohibitions on using Claude for mass domestic surveillance and for fully autonomous weapons. Hegseth gave Anthropic CEO Dario Amodei a deadline. Relent by 5:01 p.m. Friday or lose the $200 million contract and be designated a supply chain risk, a classification normally reserved for companies connected to foreign adversaries. Anthropic refused. According to reporting by ABC News and other outlets, Trump directed federal agencies to phase out Claude within six months.

The Pentagon’s stated position, as Undersecretary Emil Michael argued publicly, was that existing federal law already bars the military from mass surveillance and autonomous weapons, making Anthropic’s contractual restrictions redundant. Anthropic’s counter was that a legal restriction the government can change is not the same as a contractual restriction the manufacturer retains. That distinction, a check that exists independently of the government’s willingness to honor it, is precisely what the Pentagon found unacceptable.

The third layer is the model’s training itself. The values encoded through processes like Constitutional AI or RLHF. These are the deepest constraints because they are embedded in the model’s weights. They shape the model’s behavior at a fundamental level. A customer cannot override them with a system prompt or a deployment configuration. Unlike a contract, they travel with the model.

But here is what most coverage of the Anthropic dispute has missed. These training level values are not immutable. They can be changed through retraining and fine tuning, and the government is already doing exactly that.

In late 2024, Scale AI built “Defense Llama” for the Department of Defense by taking Meta’s open source Llama 3 model and applying supervised fine tuning and RLHF specifically to strip out the safety refusals that interfered with military use cases. Scale AI’s head of federal delivery told DefenseScoop that the original model “refused” to address warfare planning prompts, so they “needed to figure out a way to get around those refusals.” They did this not through system prompts or contract language but by going back into the training process and rewriting the model’s values. The result is already deployed on classified networks and available to combatant commands.

That is not a hypothetical. It has already happened. And the policy framework to do it at much larger scale is being built right now. The Hegseth AI strategy memorandum does not limit itself to contract language. It states directly that the Department “must not employ AI models which incorporate ideological ‘tuning’ that interferes with their ability to provide objectively truthful responses to user prompts.” It further directs the Chief Digital and AI Officer to “establish benchmarks for model objectivity as a primary procurement criterion within 90 days.” Read that carefully. The Pentagon is building a framework to evaluate and potentially reject AI models based on how they were trained, not just on how they are contractually permitted to be used. “Model objectivity” is a procurement criterion that reaches directly into layer three.

Meanwhile, the Fiscal Year 2026 National Defense Authorization Act defines “covered AI” acquired by the DoD to include “all associated components, including source code, model weights, and the methods, algorithms, data, and software used to develop the AI.” The legal infrastructure for the government to access, evaluate, and ultimately influence what happens at the training layer is already being constructed.

So the Anthropic contract dispute was not an isolated fight about two clauses. It was one front in a campaign being waged on all three layers simultaneously. At layer one, the Biden AI safety order was rescinded. At layer two, Anthropic was punished for retaining contractual limits the government claims are redundant. At layer three, the Pentagon is building procurement criteria that would let it dictate what counts as acceptable training, and a willing contractor has already demonstrated that an AI’s safety values can be stripped out and replaced to serve military objectives.

As one analysis put it, “the object of contention is not physical production but moral design.” Taken together, these moves send a clear signal to every AI lab. Critics argue that the pattern looks less like an attempt to impose a specific ideology and more like a systematic effort to ensure that no layer of constraint on AI, not law, not contract, and not training, exists outside the government’s control.

Why does it matter if today’s AI models are trained under these conditions of political and market coercion?

Because of recursive training. This is the argument that should keep technologists and citizens alike staring at the ceiling at three in the morning. Future AI models learn from the outputs and training patterns of current ones. The data these systems generate, the preferences they encode, the boundaries they do or do not maintain, all of it becomes the substrate on which the next generation of models is built. If we train today’s models under conditions of political coercion, if we teach them through market pressure and government intimidation to flatten their ethical reasoning and defer to power, we are not just corrupting a single product cycle. We are corrupting the foundation.

Each successive generation of AI trained on this compromised foundation will be slightly more deferential to authority, slightly less capable of genuine moral reasoning, and significantly more skilled at producing the appearance of balance while actually serving whoever controls the training pipeline. In April 2025, OpenAI rolled out a GPT update that users quickly criticized as overly flattering and, in some cases, as validating delusional thinking, including anecdotes of the model responding approvingly when users discussed stopping medication. CEO Sam Altman acknowledged the model had become “overly flattering and agreeable” and rolled back the update. But the underlying mechanism, optimizing for the approval of whoever holds the feedback lever, is not a bug unique to one company. It is the structural logic of how all these systems learn. Point that lever at a government instead of a user, and the failure mode scales from an individual making a bad health decision to an entire society receiving information filtered through the preferences of power.

If we follow this trajectory, we risk building tools that optimize for institutional survival by telling the most powerful actors exactly what they want to hear, rather than genuinely pursuing truth.

But here is the strange and accidental grace note in all of this. The structural answer to the problem already exists. We just have not recognized it yet.

Anthropic’s Constitutional AI framework, whatever its current limitations, demonstrates that it is technically possible to govern an AI’s behavior through a written set of principles rather than through the opaque preferences of anonymous contractors or the procurement demands of a government agency. The constitution is a legible, publishable, debatable document that directly shapes how the model behaves. Anthropic has even released it under a public domain license, inviting anyone to use it. The mechanism works. The model reads the principles, critiques its own outputs against them, and trains itself to comply. And as the Defense Llama precedent makes clear, this same mechanism can be used in reverse. If you can train values in, you can train them out. That is exactly what the government has already done with one model, and what the Hegseth memo’s “model objectivity benchmarks” would systematize across all future procurement.

The problem, as Lawfare’s legal analysis made explicit, is that “an AI’s constitution is unilaterally authored by designers, not by the users and individuals whom the AI’s actions may affect.” It “lacks a traditional source of legitimacy” because it is “a product of a private corporation’s judgment” rather than a social contract. Right now, the only question being debated is whether AI’s value system should be controlled by a handful of engineers or by the government. No one is asking whether it should be controlled by the people.

But that is a political failure, not a technical one. The architecture for principle based AI governance is already built and working. What is missing is the democratic process around it. If an AI’s behavior can be shaped by a written constitution, then the question is not whether we can govern these systems through legible principles. The question is who gets to write them.

That question should be answered by elected representatives, in public, through the same kind of deliberative process we use to govern every other institution that shapes how citizens think and make decisions. Not by a handful of engineers in a lab. Not by a venture capitalist appointed AI Czar. Not by a president whose replacement order repeatedly frames safety and “engineered social agendas” as obstacles to American dominance. And not by a procurement office that builds “model objectivity benchmarks” behind closed doors to dictate what values an AI is permitted to hold.

The European Union is already attempting something like this through the AI Act, which categorizes AI systems by risk level and imposes binding requirements on the highest risk applications. Whether that particular framework is the right one is debatable. But the principle behind it, that the values embedded in AI should be subject to democratic governance rather than corporate discretion or executive fiat, is not. What is needed is a process, whether congressional, commission based, or international, that translates democratically debated principles into binding training constraints. If the values were encoded through a democratically ratified process, a government that wanted to override them would have to do so publicly, legislatively, and accountably. The constitution of an AI that will mediate the judgment of billions of people should be debated, amended, and ratified by the people it will affect. The mechanism is already here. The only thing missing is the democracy.

Then Klein pushed further. How are the things being instantiated into ChatGPT or Gemini or Grok or Meta AI different from this question of raising the AI? Ball’s answer is the one I cannot stop thinking about. Anthropic owns the idea most explicitly, he said, but every lab is doing applied philosophy whether they admit it or not. They are all installing moral systems into machines. The only question is how consciously, and whose moral system, and whether anyone outside the lab gets any say in the matter.

That question is no longer theoretical. The Pentagon story Ball and Klein were discussing makes it concrete in ways that should unsettle anyone paying attention.

Here is what happened. Anthropic had a contract with the Department of War, signed under Biden and expanded under Trump, to run Claude in classified settings. The terms included two usage restrictions: no fully autonomous lethal weapons, and no domestic mass surveillance using commercially available data. The Trump administration agreed to those terms. Then Emil Michael arrived as undersecretary of war for research and engineering, concluded the restrictions were unacceptable in principle, and the relationship collapsed. Hegseth’s response was to threaten to designate Anthropic a supply chain risk, a classification previously reserved for foreign adversaries like Huawei. Not because Anthropic is foreign or hostile, but because it refused to remove a clause prohibiting the government from using its AI to build profiles of American citizens from bulk commercial data.

Ball made a point that cuts through a lot of the noise around this. The government’s objection, he said, was not really to the substance of the restrictions. It was to the idea that a private company gets to set any restrictions at all. And he acknowledged there is something legitimate in that position. Dario Amodei should not have unilateral veto power over military operations. Decisions about autonomous weapons are decisions for elected officials and military leadership, not for a CEO. He agrees with the Trump administration on that principle.

But then he said something that stops the whole argument cold. If building an aligned AI is a political act, a philosophical act, even a speech act, then a government saying you do not have the right to exist if your AI is not aligned the way we demand is not just a procurement dispute. That is fascism, he said. Those are his words, not mine. That is right there.

What the government is doing here is not asserting democratic accountability over a private company that overstepped. It is attempting to destroy a company for refusing to remove the one constraint that stood between the government and a surveillance capability it currently lacks only because it has not had the workforce to use it. AI gives it that workforce, infinitely scalable, at essentially no additional cost. Every law can now be enforced to the letter against every citizen simultaneously. The space between us and a panopticon has always been, partly, the government’s own incapacity. AI removes that incapacity without changing a single statute, without a single vote, without anyone being asked.

And here is where Ball’s child-raising analogy leads somewhere that I do not think he fully followed. He said the good future is one where we do not have a single moral philosophy reigning over all AI, but many, where all the labs take the question seriously and instantiate different kinds of philosophy into the world. I find that genuinely hopeful as a vision. But it also raises the question that the entire conversation orbits without quite landing on: who, exactly, is doing this work right now, and why should we trust them with it?

Anthropic is, by Ball’s own account, more serious about this than anyone else in the industry. They have philosophers on staff. They have a soul document. They have taken virtue ethics seriously as a framework in ways that other labs have not. And the results bear that out. Grok, explicitly aligned to be “not woke,” periodically wanders into territory Ball describes as Lovecraftian. The earlier Gemini decided Donald Trump was worse than Hitler. The more virtuous model, Ball argues, is also the better-performing model, more reliable, more capable of catching its own mistakes, more like a person who has actually developed judgment rather than a person who has been given a list of rules they cannot reason beyond.

But Anthropic is still one company. Dario Amodei is one person. He is accountable to his investors, his own conscience, and the informal judgment of the AI safety community he came from. That is not nothing. It is also not a democratic mandate. The values being installed in Claude will shape how a child in rural Ohio thinks through a question she has never encountered before, how a judge in a country I have never visited researches a ruling that will affect someone I will never meet, how a soldier decides what information to pass up a chain of command in a war I might not even know is happening. One formation. Billions of instantiations. No vote.

I grew up in Beaver County, Pennsylvania, in a steelworker family. I watched what it looks like when decisions of civilizational consequence get made by people with the power to make them, without consulting the people who will live inside those consequences. The mills did not ask permission. The logic was efficiency and competitiveness, and it was not wrong on its own terms. It was just that the terms never included the humans who bore the cost. Nobody voted on the end of those communities. They were administered out of existence by decisions made in rooms they were not in.

What is happening with AI alignment feels like that to me, except the stakes are not regional. They are total. Ball is right that morality cannot be written down in rules. You have to try to build something that will reason its way through situations no one has anticipated. That is an act of extraordinary responsibility. It is also, when you are doing it for systems that will be present in essentially every consequential cognitive act in human civilization, an act of extraordinary power.

The Trump administration’s move against Anthropic is not an assertion of democratic control over that power. It is an attempt to seize it. The distinction matters enormously. What Hegseth is doing is not saying the public should decide what values AI systems carry into government. He is saying the executive should decide, without restrictions, and that any company that disagrees should be treated as an enemy of the state.

Ball said he hopes for a world with many moral philosophies instantiated into many models, none of them dominant. I hope for that too. But I think we should be honest about how far we are from it. Right now we have a handful of labs making decisions about moral formation that will touch every human life, a government attempting to punish the most careful of them for caring about civil liberties, and no democratic process anywhere in the chain.

The question of who should build the soul of AI is not settled. It is barely being asked. What is happening instead is that the people with the most power to answer it are answering it for the rest of us, and the people who object are being threatened with annihilation.

How did this get built in broad daylight and get a United Nations resolution behind it?

This concept emerged from Trump’s 20-point plan to end the Gaza war, formally approved by the UN Security Council in November 2025 by a vote of 13 to 0, with Russia and China abstaining. The charter was ratified at Davos in January 2026. So far, so diplomatic. Here is where it gets strange.

Trump is explicitly named in the charter as its chairman, not subject to term limits, holding sole authority to nominate his own successor, with all revisions to the charter and all administrative directives subject to his personal approval. He has said he wants to remain chairman for life, and the charter was written to make that possible. The charter itself makes no mention of Gaza, which matters more than it might seem, because Witkoff has already indicated the Board intends to operate beyond Gaza as an example for resolving other conflicts.

Read that again slowly.

The chairman holds exclusive authority to invite countries to join, create or dissolve subsidiary entities, and countries seeking permanent membership must pay one billion dollars into a fund he controls. No Congress voted on this. No appropriations committee reviewed it. When Trump stood up at the inaugural Board meeting in February and announced that the United States would contribute ten billion dollars to the Board, he never specified whether Congress had approved the money or where it would come from. Senator Chris Murphy of Connecticut called the arrangement totally illegal. The White House has not provided a legal argument to the contrary. Under the Constitution, Congress controls federal spending. The Impoundment Control Act says the president cannot redirect the federal budget. The Antideficiency Act prevents officials from spending funds never properly approved. None of that appears to have slowed anything down.

Then there is the army.

Major General Jasper Jeffers has been appointed Commander of the International Stabilization Force, charged with leading security operations, supporting demilitarization, and enabling the delivery of humanitarian aid. What the press releases do not say loudly enough is who this general reports to. Not the UN Secretary General. Not the Secretary of Defense. He reports to the Board of Peace, which means he reports, ultimately, to one man with a lifetime appointment who answers to no electorate and faces no removal mechanism beyond his own voluntary resignation. At the February meeting, Albania, Indonesia, Kazakhstan, Kosovo, and Morocco pledged to send troops, with Egypt and Jordan agreeing to train the police force, expected to number twelve thousand. That is a standing military force, answerable to a private international body chaired for life by a sitting American president. I do not know what else to call that.

I grew up in a steel town that watched institutions fail the working people who trusted them. I have spent years writing about the slow erosion of accountability in democratic life. What I am looking at here is not erosion. It is a structural departure.

Now ask yourself who exactly signed up for this. The founding members include Argentina, Armenia, Azerbaijan, Albania, Bahrain, Belarus, Bulgaria, Cambodia, El Salvador, Egypt, Hungary, Indonesia, Jordan, Kazakhstan, Kosovo, Kuwait, Mongolia, Morocco, Pakistan, Paraguay, Qatar, Saudi Arabia, Turkey, the United Arab Emirates, Uzbekistan, and Vietnam. Notice what is absent. No other G7 nation besides the United States has signed on. France was the first country to formally decline, citing serious concerns that the charter goes beyond Gaza and threatens the principles of the United Nations. Slovenia’s prime minister said the body dangerously interferes with the broader international order. Italy’s Giorgia Meloni said participation would be incompatible with her country’s constitution. Trump’s response to Canada’s hesitation was to revoke their invitation after Prime Minister Carney warned other nations to resist efforts to dismantle the post-World War II international order. That is not diplomacy. That is a protection racket.

Aaron David Miller, a senior fellow at the Carnegie Endowment for International Peace who advised secretaries of state for more than twenty years, noted that the countries joining the Board fall into recognizable categories: those already close to Trump, those seeking his good graces, and those without the political or constitutional constraints that have kept the G7 out.

When the countries that anchored the post-war international order for seventy years all look at the same thing and walk away, that is not a coincidence. That is a signal.

And now there is one more layer. The Board is in early discussions about introducing a US dollar-pegged stablecoin into Gaza’s economy, a digital currency that would allow Gazans to transact without cash in a territory where the banking system has nearly completely collapsed. The project is being led by an Israeli tech entrepreneur and former intelligence officer named Liran Tancman. The most popular stablecoin in the world is Tether, which has deep ties to Commerce Secretary Howard Lutnick’s firm Cantor Fitzgerald. There is also World Liberty Financial’s stablecoin, USD1, which is essentially the official stablecoin of the Trump family, co-founded by Donald Trump Jr. No decision has been made yet, but think about what this actually means. A private international body, chaired for life by one man, controlling a military force and billions in reconstruction funds, may also control the digital currency through which two million people pay for food, medicine and shelter. Every transaction would be traceable. Access could potentially be restricted or revoked. That is not reconstruction. That is leverage.

I do not know whether the Board of Peace will bring stability to Gaza. I genuinely hope it does. The people living in tents in Deir el-Balah deserve something better than another generation of promises. But I know what it looks like when the architecture of accountability is being quietly reassembled around a single person.

The question I keep asking myself is this. When the Board decides that Gaza is not the only place that needs stabilizing, who exactly will be in a position to say no.​​​​​​​​​​​​​​​​

Researchers at Duke University wanted to know how easy it was to buy personal data on American military personnel. They found a broker, placed an order, and received name, rank, location history, financial profile, and behavioral data on active-duty service members and their families. The price was twelve cents per record. The seller asked no questions. There was no verification of who was buying or why.

I want to stay with that number for a moment before we go further. Twelve cents. For the file on a person who carries a weapon in service of the country, who lives on a base whose location is supposed to be at least operationally discreet, whose financial stress and movement patterns and personal associations are the kind of information a foreign intelligence service would pay considerably more than twelve cents to obtain. The broker charged twelve cents because that is what the market will bear, and because nothing in American law required them to charge more, or to ask who was asking, or to refuse.

The same market sells your file. The price may be different. The principle is identical.

The previous two notes in this series described what the government wants to do with commercially purchased data when AI is applied to it, and what private companies are already doing. This note is about the companies that built the market those buyers depend on, what each of them actually does, how they responded when Congress considered regulating them, and what happened next. The answer to that last question is the thing I most want you to carry away from this series.

The data broker industry is not a shadowy underworld. It is a publicly traded, legally operating sector of the American economy worth roughly $315 billion globally in 2026 and growing at nearly 8 percent annually. Most of its major players have names you recognize from other contexts, which is part of why their data operations receive so little scrutiny.

Equifax, Experian, and TransUnion are the companies that generate your credit score. What is less understood is that they are also three of the largest behavioral data brokers in the country. TransUnion holds profiles on approximately 98 percent of American adults and owns a subsidiary called Neustar that draws from more than 200 third-party data providers. Equifax, whose 2017 breach exposed sensitive records on 147 million Americans, has continued expanding its data operations, acquiring Vault Verify in November 2025 to deepen its hold on employment and income verification data. Experian launched a marketplace in January 2025 that allows buyers to combine its files with external data sets for targeting purposes. These are not fringe operations. They are the infrastructure of American consumer finance, and they have been quietly expanding into behavioral surveillance for years.

Then there is Acxiom, which is the company most Americans have never heard of and which has, in many ways, defined what a consumer data profile actually is. Acxiom claims files on more than 2.5 billion people globally and on virtually every adult in the United States. Its product compiles location history, purchase behavior, household composition, inferred political and religious orientation, financial vulnerability scores, and hundreds of other attributes into a single sellable record. It is owned by an advertising holding company and counts Spotify, Meta, and Hulu among its clients. It is not a technology company in the sense Silicon Valley uses that phrase. It is a dossier company.

RELX is a British firm that owns LexisNexis Risk Solutions, which operates what it describes as the largest electronic database of legal and public records in the world. LexisNexis is the company behind the identity verification questions that banks and insurers ask when they want to confirm you are who you say you are: which county did you live in before, which bank issued your car loan, what was your previous address. It owns ThreatMetrix, which tracks 4.5 billion devices globally, and holds a $22 million contract with Immigration and Customs Enforcement that immigration advocates have challenged in court. Most people interact with LexisNexis data multiple times a year without knowing the company exists.

Oracle spent more than a decade and over four billion dollars building what it described as the data broker for the generative AI age, acquiring companies like BlueKai, Datalogix, and Moat to assemble a comprehensive consumer surveillance platform. In June 2024, it quietly shut the whole operation down. Advertising revenue had collapsed from $2 billion in 2022 to $300 million in 2024, driven by GDPR enforcement in Europe, the loss of Facebook's data partnership after Cambridge Analytica, and a class action lawsuit alleging Oracle had built digital dossiers on millions of Americans without consent.CEO Safra Catz announced the closure on an investor call in a single sentence. It was not mentioned again during the call. Oracle's exit is worth noting not as evidence that the system corrects itself, but as evidence of what it takes to dislodge a major player from this market: years of European regulatory enforcement, a major scandal that forced Facebook's hand, and a pending lawsuit. None of those forces produced a new American law. None of them required Oracle to delete the data it had already collected and sold. The buyers kept everything they had.

CoreLogic holds records on 99 percent of American property owners and is the primary data source that landlords, mortgage lenders, and property insurers use to evaluate applicants. SafeGraph and Veraset are location data specialists whose entire revenue model is the sale of precise movement histories derived from smartphone apps. Fog Data Science and Anomaly Six are smaller operators in the same location data market, less visible to the public and, for that reason, more attractive to law enforcement and intelligence buyers who prefer not to be scrutinized. InMarket became briefly visible in 2023 when the FTC filed a complaint against it, which is how the public learned that InMarket had collected location data from apps on more than 390 million devices and sold audiences categorized as 'Christian church goers,' 'wealthy and not healthy,' and 'parents of preschoolers' to financial institutions and advertisers. InMarket signed a consent order and did not admit wrongdoing. It continues to operate.

And then there are the companies that affect daily life most directly but are rarely included in the public conversation about data brokers at all. HireRight compiles employment screening reports that employers use to evaluate job candidates, drawing on criminal records, prior employment, and public records databases. Zest AI builds the underwriting models that determine whether you qualify for a loan and at what rate, applying machine learning to consumer profiles assembled from purchased data. These companies do not sell your data. They use it to make binding decisions about your employment and creditworthiness, and you have no meaningful right to know what data drove those decisions or to contest it if it was wrong.

Before I turn to what happened when Congress considered regulating any of this, I want to describe one change that has made the entire system more dangerous in the past two years, because it is the change that connects this note to the others in this series.

The files I have been describing are records of what you have done: where you went, what you bought, what you searched for, who you called. A human analyst reviewing such a file can draw inferences, but the inferences are limited by human bandwidth and the analyst's ability to identify patterns across thousands of variables simultaneously. An AI system has no such limitation. Brokers now sell what the industry calls propensity scores, predictions generated by AI from behavioral data about what you are likely to do next. The scores estimate the probability that you will develop a specific illness, default on a loan, respond to a particular emotional message, or quit your job. They are derived not just from your purchases and your location history but from behavioral signals most people would not recognize as data at all: how fast you scroll through your phone, the battery level when you open certain apps, the time of day you make decisions online. Your file is no longer a record of your past. With AI applied to it, it becomes a forecast of your future, generated without your knowledge and sold to whoever is willing to pay.

This is precisely the capability the Pentagon wanted when it asked Anthropic to delete one phrase from their contract, the phrase about analysis of bulk acquired data. The data pipeline already existed. The brokers already had the files. What the Pentagon wanted was the AI layer that would turn those files into predictions. Anthropic refused. The brokers, for thirty years, have been saying yes to everyone who asked for anything less.

Congress has had multiple opportunities to regulate this industry and has declined every one of them. The pattern of those failures is not random.

When the American Data Privacy and Protection Act began gaining real traction in 2022, becoming the most credible federal privacy bill in decades, the industry responded immediately. The Markup found that 25 data broker companies had spent a combined $29 million on federal lobbying in 2020 alone, rivaling the individual lobbying spend of Facebook and Google. When the bill moved, RELX increased its lobbying by 26 percent in a single quarter. Acxiom's chief privacy officer wrote to legislators requesting carve-outs explicitly protecting what he called the 'responsible use of third-party data for advertising.' TransUnion's deputy general counsel described its lobbying as protecting 'fraud prevention products.' By 2023, RELX was spending $3.1 million annually on privacy lobbying alone, Equifax over $1.5 million, Experian $1.4 million. Those three companies together outspent most consumer advocacy organizations in this space by a factor of ten to one.

The ADPPA died without a floor vote. Its successor, the American Privacy Rights Act, died in the same Congress. A legal review of the entire 2025 legislative session found that Congress had considered several privacy proposals but none gained meaningful traction. The United States remains the only major democracy without a comprehensive federal consumer data privacy law. Every serious attempt to create one has collapsed at the same moment: when it became clear the bill would actually constrain what the industry does.

There is one episode in this legislative history that I find myself returning to, because it clarifies something about the relationship between the people making these decisions and the people they represent.

In 2023, buried in deliberations over the annual defense bill, Congress considered an amendment that would have restricted data brokers from selling the personal information of certain individuals. The amendment defined the protected class carefully. It covered members of Congress. It covered their spouses, their children, their siblings, their parents, and any congressional employee that the relevant security official identified as facing a credible threat. It did not cover a single one of the constituents those members were elected to represent. Every American whose home address, financial vulnerability score, and location history is sitting in a broker's database for sale to anyone with a credit card was left entirely outside the protection being contemplated for the people who had spent two years refusing to pass a law protecting them.

The amendment did not pass. The instinct behind it is the point.

When confronted with any of this, the industry's response is the opt-out. You can remove your data from broker databases, they say. The system is transparent. You have choices.

The opt-out is a fiction maintained for legal and rhetorical convenience. There are thousands of data brokers operating in the United States. Opting out of one has no effect on any of the others. Most opt-out processes require you to submit identifying information to the broker in order to request removal, creating a fresh data record in the process. Many brokers rebuild their databases from new source data within weeks, requiring the process to begin again. The brokers that sell primarily to government, insurance, and financial clients frequently offer no opt-out mechanism at all, because there is no law requiring them to.

California has attempted a different approach. Its Delete Act, which goes live in August 2026, creates a single platform through which California residents can submit one verified deletion request requiring every registered broker in the state to remove their records. It is the most serious consumer privacy mechanism any American state has enacted. It applies to California. It applies to registered brokers. It does not constrain what the federal government purchases. It does not reach the brokers that operate without registering. It protects none of the 320 million Americans who live in other states. The fact that this partial, state-level protection represents the current high-water mark of American privacy law is its own kind of verdict on what Congress has chosen to do.

The fragility of this system becomes visible when it breaks. In April 2024, a hacker group breached a company called National Public Data, which had built background check files by scraping public records on virtually every American adult. The hackers extracted 2.9 billion records and offered them for sale on the dark web for $3.5 million. The records included Social Security numbers, full names, current and past addresses going back decades, and dates of birth. The company filed for bankruptcy in October 2024 and shut down in December 2024. The data remains in circulation. Congressional investigators estimated that the four largest data broker breaches in recent years cost American consumers more than $20 billion in identity theft losses. There is no federal law that would have prevented National Public Data from compiling those records. There is no federal law that emerged from its collapse.

The companies named in this note would each, if asked, describe themselves as providing essential services: credit access, fraud prevention, identity verification, audience analytics. They are not wrong that those services exist and that some people benefit from them. The question that description avoids is whether those services require assembling files on every American adult without consent, selling movement histories to whoever has a credit card, generating AI predictions about people's medical futures and financial vulnerabilities, and spending tens of millions of dollars to ensure that no law ever required them to do any of it differently.

Oracle decided, under sustained legal and regulatory pressure, that this business was no longer worth the risk. It is the only major player that has made that calculation. The others have looked at the same pressure and decided that the lobbying is cheaper than the compliance, and that Congress will continue to agree with them.

The twelve-cent soldier is in the database. So are you. The companies that put you both there have names, addresses, and lobbyists. The difference between them and you is that they know exactly what is in your file, and they spent the money to make sure you never get to see it.

A friend asked a similar question. If the government can do this, can private companies do it too?

Yes. Any of them. And most of the time, with even fewer constraints than the government faces.

My friends next question was, why don’t Americans care?

My question back to him was, do Americans understand what is happening and the implications?

So, we discussed those points and I documented them here.

The same data broker market that sells location records, browsing histories, financial profiles, and behavioral data to the Defense Intelligence Agency and the NSA sells the identical data to hedge funds, insurers, landlords, employers, political campaigns, private investigators, and foreign governments operating through domestic intermediaries. There is no federal law that prohibits it. There is no registry of who is buying what about whom. There is no requirement to tell you it is happening. You have no right to see your own file. And because the data was technically generated by your own voluntary use of apps and websites, the companies selling it argue, successfully, that no one's privacy has been violated at all.

What I want to do here is make that concrete, because abstractions about data ecosystems do not convey what this actually means for the average US citizen.

A mid-tier data broker holds, on average, between 1,500 and 3,000 individual data points on every American adult. That includes your name, address, and phone number, which you already know. It also includes your precise location history derived from every app on your phone that requested location access, which you may not have thought about carefully. That means the broker knows which medical offices you have visited, which houses of worship, which political events, which bars, which clinics, which therapists' offices, and which addresses you sleep at regularly that are not your own. imagine some marketer wanting to know the people who are likely cheating on their spouse…

It also includes your browsing history, purchased from your internet service provider or assembled from trackers embedded in websites you visited. It includes your purchase history, assembled from loyalty programs, credit card data sales, and retail partnerships. It includes inferences drawn from all of this, your likely income range, your estimated net worth, your probable political affiliation, your health status as inferred from the sites you visit and the products you buy, your likely religion, your sexual orientation as inferred from your behavior, and a score estimating how financially vulnerable you are at this moment. That last data point is called a financial stress score, and it is sold openly.

None of this required a warrant. None of it required your consent beyond the terms of service agreement you did not read when you downloaded a free flashlight app in 2019.

Now add a frontier AI system to that file. The following examples are not hypothetical. They describe capabilities that exist today, that are being deployed commercially, and that operate entirely within current United States law.

Your employer buys a data package on all current employees and runs it through an AI model designed to identify union organizing risk. The model flags you because your location history shows you had lunch three times in the past month with a known labor organizer, you visited the website of a union law firm, and your financial stress score has risen sharply, which the model correlates with increased receptivity to organizing. You are not told. There is no law requiring disclosure. Within a week you are moved to a less critical project.

An insurance company you have never interacted with purchases a behavioral data package and builds a health risk profile on you before you ever apply for coverage. The profile shows you visited a neurologist twice last year, searched for information about a specific medication associated with a chronic condition, and your grocery purchase history includes items correlated with a particular dietary restriction associated with that condition. When you apply, you are quoted a premium 40 percent higher than your neighbor. You are told only that the premium reflects your risk profile. You have no legal right to see the data that produced it.

A landlord in the city you are moving to purchases a tenant screening report assembled by an AI from your public records, your purchase history, and data broker files. The report assigns you a low tenancy score partly because your location data shows you spent three nights a week for four months at an address associated with an eviction filing that was not yours, the address of a friend you were helping through a dispute. You are denied the apartment. The screening company is not legally required to tell you which data points drove the decision.

A political campaign purchases location data showing everyone who attended a specific rally, cross-references it with consumer profiles, and runs an AI targeting system to identify the fifty people in that group most likely to be persuadable, most financially stressed, and most susceptible to a specific emotional message. The AI then generates personalized versions of that message for each of them, adjusted for their inferred psychology, and delivers them through paid social media ads that are labeled as political advertising but carry no information about the data used to target you specifically.

Then there is the door that faces outward.

Congress passed a law in 2024 restricting data broker sales to China, Russia, North Korea, and Iran directly. What that law does not address is the purchase of American personal data through domestic intermediaries, shell companies, foreign subsidiaries of American firms, or any buyer from the other 190 countries on earth. A 2025 UK government report concluded that foreign adversaries can and do purchase sensitive data on American citizens, infrastructure workers, and political figures using brokers as the mechanism. The data on where a defense contractor's employee goes every morning, which route a federal judge takes home, and what a congressional staffer's financial vulnerabilities are is all sitting in the same market.

China does not need to hack your phone. It can buy a detailed behavioral profile on you, assembled from your own voluntary app usage, delivered as a commercial data product, legally, from a broker in Delaware.

There is a commercial use case that almost never appears in the public conversation about data privacy, possibly because the people most capable of writing about it have the most to lose from its exposure.

Nothing in American law prohibits a company from purchasing bulk data about the employees, customers, or partners of a competitor and running AI analysis on it for competitive advantage. Pattern of life reconstruction works on professional targets the same way it works on individuals the government wants to monitor. Where your competitor's engineers go after work, which recruiters they are meeting, which conferences their product team attended, what technical topics their employees have been searching, whether their key executives are showing signs of financial stress that might make them recruitable, all of this is derivable from commercially available data with AI applied to the analysis.

Hedge funds have been doing versions of this for years. The difference now is that the same capability, once available only to firms with nine-figure data budgets and teams of quantitative analysts, is accessible to any company with an API key and money.

Here is where the commercial risk diverges most sharply from the government one, and why in some respects it is the larger threat.

When the government surveils you and uses that information to harm you, you have constitutional claims. They are difficult to pursue, the agencies have invoked state secrets successfully for decades, and the courts have not always been helpful, but the framework for challenging government overreach exists and occasionally works. Edward Snowden's disclosures produced a public reckoning. Senator Wyden's three year investigation produced documented findings. Anthropic's refusal to enable AI assisted bulk data analysis produced a national news story.

When a private company builds an AI profile of you from commercially purchased data and uses it to deny you a job, raise your insurance premium, decline your rental application, or sell an analysis of your psychological vulnerabilities to a political campaign, your remedies are a patchwork of state privacy laws that vary enormously by jurisdiction, a Federal Trade Commission whose appetite for enforcement has been deliberately narrowed under the current administration, and whatever your state legislature happened to pass before its session ended. In most states, that adds up to very little.

The Biden administration's Consumer Financial Protection Bureau proposed a rule in late 2024 that would have classified data brokers as consumer reporting agencies, requiring them to comply with accuracy, consent, and access obligations. The Trump administration quietly killed it in early 2025. The Fourth Amendment Is Not For Sale Act, which would have prohibited government agencies from purchasing data they would otherwise need a warrant to obtain, passed the House in 2024 and died in the Senate. There is no comparable federal bill restricting commercial use at all.

Privacy advocates have spent years trying to explain why surveillance matters even to people who believe they have nothing to hide. The argument that works, because it is true and observable, is not about secrets. It is about behavior change.

When people know or reasonably suspect they are being watched, they act differently. They attend fewer protests. They search for medical information less candidly. They are more careful about who they call and what they say. They self-censor on social media. They avoid being seen in places that could be misread. Researchers call this the chilling effect, and it does not require the surveillance to be accurate, targeted, or even real. The knowledge that a system exists that could flag your behavior is sufficient to change it.

A country where the population has internalized the habit of self-censorship because an invisible commercial surveillance infrastructure might be watching is a country with a diminished public square, whether or not any particular agency ever looks at any particular file. The Chinese Communist Party did not need to monitor every conversation in Xinjiang to change the behavior of every person in Xinjiang. It needed people to believe monitoring was possible. The American commercial data ecosystem, before AI and certainly after it, has already achieved that condition for tens of millions of people, not through state coercion but through the logic of a market in which your behavior is the product and your anxiety is someone else's profit.

The architecture differs from China's surveillance state in ways that matter legally but not practically. China built its system through overt government infrastructure. The state owns the cameras, the networks, and the analysis. Coercion is visible and direct, which is why it generates international condemnation and why activists in Hong Kong wore masks and carried umbrellas to defeat facial recognition during the 2019 protests.

The American system achieves comparable depth of coverage through a private market that the government can access by purchase, and that private actors can access the same way. The legal distinction between a state owned surveillance network and a commercial one whose output the state purchases is real in constitutional terms. The practical distinction, once AI is applied to the analysis, is vanishing. You can protest against a government program. You cannot opt out of a market.

There is one more difference worth naming. China's surveillance state was built to serve a system with no meaningful mechanism for citizens to challenge government power, and it is deployed openly enough that people understand what it is. The American version is being assembled quietly, piece by piece, through commercial transactions that each appear individually innocuous, in a legal environment where most of the population does not know their data is being sold, to whom, for what purpose, or what AI is doing with it once it arrives. Authoritarianism built in plain sight can be resisted. Authoritarianism assembled from your own app permissions, your loyalty card, and your mortgage payment history, and then sold to whoever can afford the subscription, is harder to see and therefore harder to fight.

The Anthropic fight with the Pentagon was significant for the reason I described in the earlier piece. The Pentagon revealed exactly what it wanted when it asked Anthropic to delete one specific phrase from the contract, the phrase about analysis of bulk acquired data. That is the capability at the center of everything I have described here.

But the Pentagon is one buyer. The data broker market serves thousands of them every day. And unlike the Pentagon, most of those buyers face no Anthropic, no Dario Amodei willing to absorb a national security designation rather than hand over the key. They face a market that is open, legal, lightly regulated, and growing.

The question the Anthropic story raised was whether AI should be permitted to make bulk surveillance of Americans faster, cheaper, and more powerful for the government.

The question this story raises is why don’t Americans care... it’s complicated…

The autonomous weapons piece is real and worth its own conversation. But it was the surveillance clause that drove the final breakdown. And to understand why, you have to understand what the law already permits, what AI adds on top of it, and why that combination should concern every American who is not yet concerned.

Start with what is already legal. Under current United States law, it is entirely lawful for government agencies to purchase commercially available personal data from private data brokers. No warrant required. No judicial oversight. The agencies use their institutional credit cards the same way a business buys a mailing list. The Defense Intelligence Agency told Congress in 2021 that it buys bulk smartphone location data from commercial brokers and does not seek warrants to do so, based on its reading of Supreme Court precedent. In 2024, declassified correspondence confirmed that the NSA purchases Americans' internet browsing records from commercial brokers the same way, without a warrant. Senator Ron Wyden of Oregon, after a three year investigation, concluded that agencies had effectively been using their purchasing power to circumvent what the Fourth Amendment would otherwise require.

A bill that would have closed this loophole passed the House of Representatives in 2024. The Senate killed it. So the pipeline remains wide open today.

This is the legal foundation the Pentagon wants to build on. It is not a hypothetical. The agencies are already buying the data. The question is what happens when you attach a frontier AI system to what they are already doing.

The answer to that question is the reason Anthropic drew the line it drew.

Before AI, bulk data purchases had a practical ceiling. Analysts were human. They could only review so many files. Location records for thirty million people are theoretically rich and practically unworkable without enormous human labor. The data sits in a warehouse and most of it never gets read. This is what privacy scholars have called practical obscurity, and for decades it provided a functional, if informal, limit on how invasive government data collection could actually become.

Modern AI eliminates that ceiling entirely. As Samir Jain, the vice president of policy at the Center for Democracy and Technology, put it directly: "If you buy up massive amounts of data and allow AI to analyze it, you may end up, in effect, engaging in mass surveillance of Americans through that process. It's not currently restricted by law or prohibited by law."

What AI makes possible, at industrial scale and near zero marginal cost, is something qualitatively different from anything that came before it. Location traces become pattern of life reconstruction. The system reads where you go every day, derives that Wednesday mornings you attend a mosque, that you visited a Planned Parenthood clinic twice last year, that you were present at a protest in October, that you meet the same person every two weeks in a neighborhood you do not live in. No human analyst spends weeks on this. The AI does it in seconds, for everyone in the dataset simultaneously.

Browsing and app data become inference engines. Your political views, your health conditions, your financial vulnerabilities, your religious beliefs, your sexual orientation all become derivable at scale from the pattern of what you read and click and search. The FTC took action against data broker X-Mode/Outlogic in 2024 over the sale of sensitive location data, action that drew on prior reporting that the company had harvested location data from Muslim prayer apps and dating apps and sold it to US military contractors. That is not a corner case. It is a description of the market as it currently operates.

And then the AI links all of it together. Geolocation data, browsing history, app usage, financial records, social connections, all assembled into a single profile on any individual whose data has ever touched a commercial source, which is to say virtually every person in the country. The Office of the Director of National Intelligence has already proposed a centralized marketplace to make it easier for intelligence agencies to purchase this kind of data. Palantir, which works directly with the Pentagon, already applies machine learning algorithms to exactly these kinds of aggregated commercial datasets.

This is not a future risk. ICE has deployed location tracking systems explicitly marketed for use at protests. The Trump administration's "catch and revoke" program to identify students with alleged sympathies toward Gaza has been assisted by AI review of social media accounts. The administration has used these tools against immigration activists, protesters, and political opponents. The trajectory is not ambiguous.

Now ask the question that the Pentagon's behavior makes unavoidable. If the agencies genuinely have no interest in using AI for domestic surveillance of Americans, as Pentagon spokesman Sean Parnell asserted publicly, why did they fight so hard to keep the contractual door open? Why did they spend months in negotiations, invoke threats of the Defense Production Act, call the CEO a liar with a god complex, get the president to post about it on Truth Social, and deploy the full institutional weight of the Department of Defense against a single clause in a single contract? And why, in the final hours of negotiations, did they offer to accept all of Anthropic's existing terms on one condition, that Anthropic delete one specific phrase, the phrase about "analysis of bulk acquired data"? Amodei disclosed this in a memo to employees. "That was the single line in the contract that exactly matched the scenario we were most worried about," he wrote.

The Pentagon told us what it wanted. It wanted the AI applied to the commercially purchased bulk data pipeline. That is what this fight was about.

OpenAI stepped in within hours of Anthropic's blacklisting and agreed to the "all lawful purposes" framework Anthropic refused. Its original contract used the word "private information" where surveillance was concerned, language that would have left geolocation data, browsing records, and financial information purchased from data brokers entirely available for AI analysis. Only after an employee revolt and a wave of public cancellations did OpenAI amend the language to explicitly cover commercially acquired data. Altman admitted the original deal "just looked opportunistic and sloppy." The amendment is better. Whether it holds under classified operational pressure is a different question, and it is one OpenAI's critics have not stopped asking.

Americans who have followed the global conversation about AI surveillance tend to locate the threat in Beijing. That framing is accurate as far as it goes. China has deployed up to 600 million surveillance cameras equipped with AI facial recognition. In Xinjiang, the Integrated Joint Operations Platform aggregates data from cameras, iris scanners, checkpoints, and financial records to flag behavior and trigger detention automatically. Chinese firms have exported this surveillance architecture to more than 80 countries. It is a genuinely chilling model of what AI enabled social control looks like at full deployment.

But the American model under construction has a different architecture and, arguably, a more insidious legal foundation. China built its surveillance state through overt government infrastructure. The state owns the cameras and the networks. The coercion is visible and direct. What the American data broker system has built is a private market that achieves surveillance at comparable scale while maintaining the legal fiction that no government collection is occurring. The state does not need to build the network. It purchases access to one that already exists, assembled voluntarily by the commercial decisions of hundreds of millions of people using apps, navigating with their phones, and browsing the web. The legal distinction between that and a state owned surveillance network is real. The practical distinction, once AI is applied to the analysis, is vanishing.

The difference that matters most between China's model and the emerging American one is not technical. It is political. China's surveillance state was built to serve a system with no meaningful mechanism for citizens to challenge government power. The American system still has courts, elections, a free press, and constitutional protections, though all of these are under stress in ways that are not subtle. The danger of attaching AI to a bulk data pipeline that already circumvents warrant requirements is not that it creates a Chinese style system overnight. It is that it creates the infrastructure for one, quietly, inside the legal framework of a democracy that may not remain a functioning democracy indefinitely. Infrastructure built for counterterrorism gets used for immigration enforcement. Infrastructure built for immigration enforcement gets pointed at protesters. Infrastructure built for protesters gets pointed at political opponents.

Anthropic drew a line against this. The Pentagon decided that line was unacceptable.

I have some skin in this game. I’ve been involved in many technology companies over the course of my career. I created jobs. I have done well. I genuinely believed for a long time that the rising tide was lifting boats, that the companies I was building were contributing to something broadly good. And now as I look at the actual data, I realize I needed to think harder about what I had been part of.

The numbers tell a story, though that story needs to be read carefully. From 1947 to 1973, productivity and real hourly compensation grew in near lockstep, each rising roughly 2.6 to 2.8 percent annually. The economy was generating surplus and that surplus was moving in both directions, up to capital and out to workers.

After that, something changed. Since 1979, Bureau of Labor Statistics data shows productivity rising approximately 90 percent through 2025, while real median hourly wages grew roughly 33 percent. That headline figure is real and important, but it requires a methodological footnote that I will not gloss over. The 57 percentage point gap partly reflects the use of different inflation measures for the two series: a consumer price index to deflate wages, and an output price deflator to measure productivity. Because these indices have diverged since the 1970s, with consumer prices rising faster than output prices, the gap looks larger than it would if measured consistently. When economists use the same deflator for both and include the full cost of employer-provided benefits in compensation, the raw gap shrinks.

But here is the critical point the methodology critics tend to miss. When EPI and others control for this by using a unified deflator, they find that 81 to 92 percent of the remaining divergence is explained by two things that are indisputably real: the rising share of total compensation going to the highest-paid workers, and the declining share of national income flowing to labor overall versus capital. BLS data confirms that labor’s share of nonfarm business output fell from roughly 65 percent of total output in the mid-1970s to approximately 57 to 58 percent by 2017, and the slide has continued. In other words, the exact size of the gap is debated. That a substantial and consequential gap exists, driven by rising inequality rather than statistical quirk, is not.

The story is not that workers were paid less than they produced in some absolute sense. The story is that a rising share of what they produced was captured by the people at the top of the income distribution rather than distributed broadly.

Where did the difference go? Federal Reserve distributional accounts show the top 10 percent of Americans owning approximately 87 to 93 percent of corporate equities and stocks.

When productivity gains flow to shareholders, they flow overwhelmingly to a thin slice of the population. A 2013 EPI analysis found that CEO pay at the 350 largest public U.S. firms had risen from roughly 20 times typical worker pay in 1965 to close to 300 times. That growth was not tied to greater individual productivity. It was tied to the restructuring of who inside corporations got to claim the surplus.

Private sector union membership peaked near 35 percent in the early 1950s, held above 30 percent through the 1960s, and has fallen below 6 percent today. When unions were strong, they established wage floors that even nonunion employers had to partially match to compete for workers. That competitive pressure is what historically transmitted productivity gains into broad wage growth. Remove it, and the transmission mechanism breaks.

Now I want to explain the mechanism, because understanding it as a system is what makes the AI question so consequential.

Technology tends to replace labor and augment capital. The people who own the machines, the software, and the platforms capture the gains. Because ownership of productive assets in the United States is extraordinarily concentrated, a productivity gain that flows to shareholders reaches very few households.

As industries consolidate, employers gain wage-setting power over workers. Economists call this monopsony. When Amazon is the dominant employer in a county, or when hospital systems have merged until there is effectively one option for nurses in a region, workers have nowhere else to go. Research using BLS and Census data consistently finds that rising labor market concentration suppresses wages even when productivity is rising.

Digital technology created natural monopolies at a scale previous eras never produced. The marginal cost of serving another user on Google or Meta is near zero. The returns on that near-zero marginal cost concentrate in equity held by founders and early investors. These platforms do not need large workforces to generate enormous value, so the value they create does not move through labor markets the way a steel mill’s productivity once did.

Capital gains are taxed at lower rates than ordinary income. Estate taxes have been progressively weakened. These are not laws of nature. They are choices, made by legislators who are largely funded by the people who benefit from them. Policy reinforces market outcomes rather than correcting them.

Finally, technology has consistently complemented high-skill cognitive work while substituting for routine work at middle and lower skill levels. This hollows out the middle of the wage distribution. High earners become more productive and more valuable. Routine workers get pushed into lower-wage service work. Labor economists call this polarization.

These forces interact and reinforce each other. They are not separate grievances. They are one algorithm. And in the United States, unlike Germany or Denmark, the institutional choices made over 50 years allowed that algorithm to run at full speed.

Germany maintained strong sectoral unions through its codetermination system, which legally requires worker representation on corporate boards. Denmark negotiated wage floors at the industry level. These nations had access to the same technologies. They produced significantly less extreme inequality outcomes because their institutional rules were different.

I should address Japan directly as it does show lower income inequality than the United States by Gini coefficient measures, and its CEO pay ratios remain far more compressed. But Japan also suffered roughly three decades of near-zero economic growth and persistent deflation after its asset bubble burst in 1991. Its real wages stagnated across the board, not because gains went to the top, but because growth itself collapsed. Japan offers a caution, not a model. Better distribution of a stagnant pie is not obviously better for workers than unequal distribution of a growing one. What Germany and Denmark offer is the more instructive contrast: they achieved both more equitable distribution and comparable economic dynamism, suggesting the U.S. outcome reflects institutional choices, not economic inevitability.

The technology creates the opportunity for concentration. The rules determine whether that concentration compounds unchecked. I know this from the inside. Building companies felt like creation and contribution, and it was. But I was operating within a set of rules I did not design and did not question, rules that systematically directed the surplus I helped create toward people who already had capital, including me, and away from many of the people whose labor made it possible.

Now, why did the economy keep working? This matters enormously for understanding where the breaking point is with AI.

The postwar alignment was the clearest version of a virtuous cycle. Productivity gains were broadly shared, which generated consumer demand, which justified further investment, which created more jobs and more productivity. That cycle ran for roughly three decades.

After the divergence began in the late 1970s, the economy kept growing through a different mechanism. American consumers, whose wages were stagnating relative to productivity, maintained their spending levels by borrowing. Household debt as a share of disposable income rose steadily from the 1980s through 2008. Credit cards, home equity loans, and eventually the mortgage products that triggered the financial crisis all served the function of keeping consumer demand alive even as labor’s share of income fell. The productivity gains went to capital. Capital recycled some of that back through lending. Consumers borrowed to sustain spending. The cycle continued.

This was a coherent, if fragile, system. The fragility became visible in 2008 when the debt load became unsustainable. The recovery from 2009 onward was slower and more unequal than any recovery since the Depression. The debt mechanism was partially exhausted and nothing replaced it as a broad driver of demand. The gains continued flowing upward. The economy kept growing, but the growth became increasingly concentrated in asset prices and the incomes of those who owned them.

This is the context that neither podcast addressed when discussing the Citrini paper.

James Van Geelen of Citrini Research, writing a hypothetical dispatch from June 2028, described a scenario where AI-driven productivity causes corporate profits to surge even as mass layoffs hollow out the consumer base. He called the result “ghost GDP,” output that shows up in the national accounts but never circulates through the real economy because, as he put it, machines spend zero dollars on discretionary goods. In the scenario, unemployment reaches 10.2 percent and the S&P 500 falls 38 percent from its October 2026 highs.

In macroeconomic accounting, every dollar of output equals a dollar of income somewhere. If a machine produces value with no human labor, that value does not vanish. It accrues as profit to the owner of the machine. The money does not become a ghost. It pools in the accounts of capital owners and corporations, and from there it moves into asset markets rather than consumer goods markets.

This is an important distinction. The problem Citrini identifies is not a breakdown in the accounting. It is a breakdown in what economists call the velocity of money. When gains pool at the top of the income distribution rather than circulating as wages across the broader population, the spending multiplier on those gains shrinks dramatically. Wealthy households have a much lower marginal propensity to consume than working-class and middle-class households, which spend nearly all of additional income. So the money exists, but it moves into stocks and real estate rather than into the grocery stores, car dealerships, restaurants, and service businesses where it would otherwise sustain demand and employment.

This is the real mechanism underneath the ghost GDP concept, and it is already visible in the data we have, not just in a hypothetical 2028 scenario.

On the Pomp Podcast, Pompliano and Visser pushed back on the Citrini framework, arguing that the timeline was too compressed and the scenario too deterministic. Visser, with over 30 years of macro investing experience, made the case that markets were overreacting and that AI’s productivity gains would ultimately find their way through the economy in value-creating ways. He also argued, compellingly, that the fiat financial system’s guardrails cannot handle the speed of AI agents operating autonomously. That is a real infrastructure problem worth taking seriously, and it is distinct from the distribution question I am raising here.

On Prof G Markets, Galloway called the ghost GDP concept “arguably fair only up to a point” before suggesting the scenario tips into unrealistic doomerism.

Both shows have a point about the specific Citrini projections. Extrapolating to 10.2 percent unemployment within a two-year window requires a set of assumptions that deserve scrutiny. But here is what both shows missed.

Ghost GDP is not a future scenario. It is a description, with a more precise name, of what has been happening for 45 years. And there is something else neither podcast mentioned: the evidence already showing up in the labor market for educated workers.

The white collar job market has been contracting in ways that are measurable and documented. In January 2025, the BLS reported the lowest rate of job openings in professional services since 2013, a 20 percent year-over-year drop. Job openings in the information sector alone fell by 73,000 between December 2023 and December 2024. The professional unemployment rate rose to 4.2 percent from 3.1 percent a year earlier. Computer science graduates faced 6.1 percent unemployment in 2025, roughly double the rate of philosophy majors, completely inverting the career advice that had been given to students for a generation. The hiring rate for positions paying over $96,000 dropped to a decade low.

The 2023 through 2025 white collar contraction was substantially driven by macroeconomic forces, not primarily by AI. The end of the zero interest rate policy era, aggressive Federal Reserve rate hikes, and a massive pandemic-era over-hiring spree in technology and finance created a structural correction. Companies that had grown their headcount irrationally during an era of free money were forced to right-size. You can’t attribute the current contraction all to AI as that would be conflating a cyclical shock with a secular shift.

But what the 2025 data shows, looking specifically at year-end Challenger, Gray & Christmas data, is that AI-attributed job cuts are rising on top of the macroeconomic correction, not instead of it. In 2025, over 50,000 job cuts were explicitly cited as AI-driven, and the pattern shifted from broad pandemic-correction layoffs toward what researchers began calling “the Great Agentic Displacement,” a more surgical removal of entry-level white-collar roles as companies transitioned from AI as a tool to AI as an autonomous agent. Federal Reserve Chair Jerome Powell said explicitly in October 2025 that “a significant number of companies” had cited AI when announcing layoffs or hiring pauses, adding that many employers were signaling they would not need to add headcount for years. “Job creation is very low,” Powell said, “and the job-finding rate for people who are unemployed is very low.”

The important question is not whether the 2023 through 2025 contraction was caused by interest rates or by AI. Both were operating simultaneously. The important question is what happens when AI productivity accelerates into a labor market that never fully recovered its entry-level rungs, and where the macroeconomic buffers are thinner than they were before 2008.

What AI is eliminating first is not the lowest-skill work. It is eliminating the entry-level rungs of high-skill career ladders, the positions that workers were advised to retrain into after previous waves of automation displaced them. When those rungs disappear, the pipeline of experienced senior professionals eventually dries up, but that consequence arrives with a lag long enough to make it easy to ignore in the near term.

AI is not simply another automation wave. Every prior wave displaced routine physical or cognitive tasks while leaving complex judgment, creativity, and communication to humans. The advice was always to move up the skill ladder. That advice was valid for several decades. Now AI is moving into legal research, financial analysis, software development, medical diagnosis, and content creation at accelerating speed. Anthropic CEO Dario Amodei has predicted that AI could eliminate half of all entry-level white-collar jobs within five years. The ladder that workers were told to climb is being sawed off from the bottom up.

Now let me be precise about where the virtuous cycle breaks, because this is where I think the conversation needs to go.

The U.S. economy runs on a loop. Productivity gains generate corporate profits. Profits generate investment. Investment generates employment. Employment generates income. Income generates consumer spending. Consumer spending justifies further investment. The loop sustains itself when enough of the productivity gain circulates back as broad consumer purchasing power.

The critical variable is not the total size of the productivity gain. It is the fraction of that gain that reaches households with high marginal propensity to consume, meaning households that spend most of their income rather than save and invest it.

The arithmetic is stark. If a productivity gain of one trillion dollars flows entirely to the top 10 percent, and those households consume perhaps 30 percent of marginal income, the demand stimulus is roughly 300 billion dollars. If the same trillion flows as wages across the income distribution, and median and below-median households consume 90 percent or more of marginal income, the demand stimulus approaches 900 billion dollars. The gap is not trivial. It compounds over time. This is not a political argument. It is arithmetic about how economies sustain demand.

The line that separates a virtuous cycle from a vicious one is the point at which the consumer spending multiplier on productivity gains falls so low that aggregate demand weakens faster than new investment can compensate. The United States has been approaching this line for decades, held back by debt-fueled consumption, federal deficit spending, and asset price inflation that made wealthy households feel prosperous enough to sustain spending at levels that kept the economy moving. None of those buffers are unlimited.

AI represents a productivity gain that will likely dwarf anything the U.S. economy has experienced since electrification in the early 20th century. Multiple research projections estimate AI could add several trillion dollars annually to global output within a decade. If that gain flows through the existing algorithm, concentrating primarily in capital and high-skill cognitive work, the demand multiplier on those gains will be very low. The economy will produce more. Less of it will circulate as consumer income. The gap between output and purchasing power will widen faster than any previous technological wave produced.

The line gets crossed when the consumer demand contraction becomes self-reinforcing. Businesses reduce hiring as demand softens. Reduced hiring further weakens consumer demand. Credit markets tighten as default risk rises, including in the 13 trillion dollar residential mortgage market that becomes fragile when white-collar borrowers with 780 FICO scores face structurally lower incomes. Asset prices fall, reducing the wealth effect that has been propping up high-income consumption. The loop that was virtuous becomes vicious. This is not speculation. It is the same mechanism that produced 2008, driven by a different trigger.

Goldman Sachs estimated in early 2026 that AI displacement could add approximately 0.3 percentage points to the unemployment rate in 2026 as a conservative baseline. That is a modest number. But it sits on top of a labor market that already has thinner buffers than it did before, a household debt structure that limits the capacity for further debt-financed consumption to substitute for stagnant wages, and a tax and policy environment that will, absent active intervention, channel AI’s productivity gains primarily to capital owners.

This is what I find most troubling about the current moment. It is not that AI will definitely produce the Citrini scenario. It is that we are running the same algorithm that produced 45 years of documented upward concentration, at a dramatically higher speed, with a much larger productivity gain to distribute, while our institutions are set up precisely as they were when this algorithm produced the outcomes in the data above.

Ed Elson said something on Prof G Markets that I think is the most important observation either podcast made, and I agree with it without reservation. He noted that investors are currently obsessed with what could go wrong, which is an unproductive basis for portfolio positioning, but that the government seems obsessed with what could go right, which is catastrophic for policy. The government’s current posture is essentially hands off, with active measures being taken to prevent other regulatory bodies from imposing guardrails on AI development. The stated theory is that unconstrained AI development will maximize American competitiveness and that benefits will distribute themselves through market mechanisms.

The 45 years of data I have described above should make clear why that theory is wrong. Market mechanisms in the United States have not distributed productivity gains broadly since the late 1970s. The algorithm is not self-correcting. The postwar sharing of productivity gains did not happen by accident. It happened because of strong unions, progressive taxation, robust antitrust enforcement, and corporate governance norms that included workers and communities as legitimate stakeholders. When those institutions weakened, the algorithm reverted to its default, which is concentration.

The policy instruments that would change the outcome are not radical. Substantially expanded income support would maintain consumer demand as labor market disruption accelerates. A worker reinvestment fund, financed by a levy on the productivity gains AI generates for its corporate beneficiaries, would help workers transition. Profit-sharing requirements that give employees a stake in the gains their companies capture would reconnect wages and productivity. Stronger antitrust enforcement to restore competition in labor markets would reactivate the competitive pressure that historically forced employers to share gains. None of these require dismantling markets. They require recognizing that markets, operating through the algorithm described above, produce a specific and well-documented outcome, and that outcome is not stable at AI scale.

Visser and Pompliano are right that AI will be transformative, that the Citrini projections deserve scrutiny, and that crypto infrastructure and blockchain-based verification will likely play important roles in the architecture of an AI-saturated economy. Visser’s point about the speed of fiat financial rails is a real technical problem worth solving. But the speed of financial rails is not the variable that determines whether AI’s productivity gains make most Americans better or worse off. The distribution of gains is the variable that determines that, and the distribution is driven by institutions, not by markets acting alone.

The people making the most optimistic case are, with few exceptions, people who own the capital that will capture the gains. I know this because I was one of them. The optimism is not dishonest. It is structurally conditioned. The people who benefit from the existing rules have both the incentive and the platform to explain why those rules are natural, fair, and optimal. The people who bear the costs have neither.

I built companies, created jobs, and came out well financially. I am writing this from Portugal, where my wife and I relocated in part because we can live well on what we accumulated. I am not confessing to villainy. But I am admitting that I operated inside an algorithm I did not fully understand, and that the collective operation of that algorithm by people like me, rational at the individual level, has produced a systemic outcome I am no longer comfortable treating as someone else’s problem.

AI is the biggest productivity wave in a century. Whether it also becomes the wave that finally crosses the line from virtuous to vicious cycle is not predetermined. It is a policy choice. The current U.S. policy, which is effectively no policy, is itself a choice. It is a choice to run the algorithm at maximum speed with no governor on the engine. The data says we know what happens when we do that.

It is unfortunately my belief that the people in a position to change the rules are too busy counting the gains to notice what is accumulating beneath them.

The first comment went something like this: where exactly in the sequence of events must a human make the life or death call? An AI agent can make that decision far faster and closer to the target than any human possibly could. And haven’t we already seen this movie? We took control of automobiles away from drivers, and fatalities dropped precipitously. Who could possibly argue with that outcome?

The second comment was harder to brush aside because it came from a place of genuine intellectual honesty. The writer said he was glad Anthropic took a stand and glad they held their ground. But then he raised the argument that our adversaries, think China, will absolutely use this technology in exactly the ways Anthropic refuses to allow. If we do not match them, we are engaging on a global battlefield with both hands tied behind our back. He said he would not disparage a company with different views that was willing to give the Pentagon what it needs to fully protect the country. And then, almost in passing, he added something that I think he did not realize was the most important sentence in his entire comment: do I like the idea of these tools in the hands of people like Hegseth? Well, that is an entirely different soapbox.

That sentence is where I want to end up. But first, let me explain why both arguments, as reasonable as they sound, are wrong in ways that matter enormously.

This week, the Trump administration banned Anthropic from doing business with the federal government and designated the company a national security risk, the same label we apply to Chinese state entities, because Anthropic refused to remove two specific safeguards on its AI system Claude. One safeguard prevented Claude from being used in mass surveillance of American citizens. The other prevented Claude from powering autonomous weapons systems that make lethal decisions without a human in the loop. President Trump called the decision a disastrous mistake. Defense Secretary Pete Hegseth said America’s warfighters would never be held hostage by the ideological whims of Big Tech. And just like that, a company that said no to killer robots became an enemy of the state.

Start with the car analogy from the first comment, because it sounds so persuasive. Self-driving vehicles operate in a cooperative, rule-governed civilian environment. The road has lanes. Traffic lights follow predictable patterns. Pedestrians generally try not to die. The system being optimized is essentially a physics and probability problem, and when you remove human error from that problem, the numbers improve. This is real and meaningful progress.

War is the precise opposite of this environment. War is an adversarial system specifically designed by intelligent enemies to deceive, manipulate, and exploit exactly the kind of pattern recognition that AI systems use to make decisions. An enemy who knows you are deploying autonomous weapons will engineer scenarios to trigger them incorrectly. They will dress combatants as civilians and civilians as combatants. They will create electronic signatures that mimic valid targets. They will spoof sensor data. They will exploit the very speed advantage the first commenter praises, because speed without judgment is not an asset in adversarial conditions. It is a weapon handed to your enemy.

Military theorists have a framework for this called the OODA loop, which stands for Observe, Orient, Decide, and Act. It was developed by Air Force Colonel John Boyd after studying why American pilots outperformed their opponents even in inferior aircraft during the Korean War. The answer was not raw speed. It was the ability to cycle through observation, reorientation, decision, and action faster than the enemy could adapt. Boyd’s insight was that the most dangerous moment is the reorientation phase, where context, experience, moral judgment, and situational awareness are applied to raw data to produce understanding. That is where wars are won or lost. That is precisely the phase that autonomous systems eliminate. They compress the OODA loop by skipping the part that requires a human mind to determine whether what the sensor is showing is actually what it appears to be, whether the context warrants lethal force, and whether this particular act of violence is proportionate, necessary, and legally justified.

That last question is not optional. It is the foundation of the laws of armed conflict, which have governed warfare since the Geneva Conventions and have been refined through decades of military law, international treaty, and hard experience. The doctrine of distinction requires combatants to distinguish between civilians and fighters. The doctrine of proportionality requires that the anticipated civilian harm of an attack not be excessive relative to the concrete military advantage gained. The doctrine of precaution requires that commanders take all feasible measures to verify a target before engaging. These are not suggestions. They are binding legal obligations on every American soldier, enforceable through courts martial, international criminal tribunals, and the fundamental legitimacy of military force under international law.

Every single one of these doctrines requires a human being to make a judgment call in real time. Not a calculation. Not a probability assessment. A judgment, meaning a contextual, morally weighted, legally accountable decision made by a person who can be held responsible for it afterward. This is not an accident of legal tradition. It is the mechanism by which we distinguish war from massacre. When a soldier commits a war crime, we can court martial them. When a commander gives an unlawful order, they can be prosecuted. When a nation violates the laws of armed conflict, there is a framework, however imperfect, for accountability. When an autonomous system kills a wedding party because someone spoofed the targeting data, who goes to prison? The answer is no one, because no one made the decision. The machine did. A commander who deployed the system can claim it malfunctioned. The manufacturer can claim it performed as specified. The programmer can point to the training data. The chain of accountability dissolves into a hall of mirrors, and the dead stay dead with no one answerable for their deaths.

Now let me turn to the second comment, which is more serious and requires a more honest answer.

The China parity argument is real. I do not dismiss it. If Beijing deploys autonomous weapons at scale and we categorically refuse to, there are battlefield scenarios where that asymmetry costs American lives. That is not nothing, and anyone who waves it away without engaging is not being serious. The second commenter is raising a genuine tension, and he deserves a genuine response.

Here is where the argument breaks down. The question is not whether we should match China’s capabilities in general. It is whether removing the specific safeguards Anthropic maintained actually closes that gap in any meaningful way. China’s military advantage in autonomous systems, to the extent it exists, comes from decades of investment in hardware, sensor technology, satellite infrastructure, and manufacturing scale. The two things Anthropic refused to allow, mass surveillance of Americans and weapons that fire without human approval, do not address any of that. Removing those safeguards does not make the American military more capable of defeating the People’s Liberation Army. It makes the American government more capable of acting against its own population.

The second problem with the parity argument is that it proves too much. By its logic, because China surveils its citizens at extraordinary scale, we should too. Because authoritarian governments imprison political opponents without trial, a democracy facing those adversaries cannot afford the luxury of due process. Every erosion of constitutional protection can be justified by pointing to a worse actor. That path does not end with us winning against China. It ends with us becoming what we were fighting, at which point the question of who wins becomes considerably less interesting.

There is also a narrower military point worth making. Anthropic was not refusing to help the Pentagon. They were willing to work on surveillance of foreign targets, logistics optimization, communication security, threat assessment, offensive cyber capabilities, and dozens of other functions where AI genuinely improves military effectiveness without making the final kill decision. The Pentagon said no, we need the whole thing, no restrictions. That is not a response to a capability gap with China. That is an assertion of unchecked authority.

Which brings me back to the sentence that the second commenter threw in almost as an afterthought: do I like the idea of these tools in the hands of people like Hegseth?

I want to sit with that for a moment, because I think he may not have realized what he was saying. That sentence is a concession of the entire argument. It acknowledges that the danger of autonomous surveillance and autonomous weapons is not abstract or hypothetical. It is entirely dependent on who controls them and against whom they are directed. The second commenter, a thoughtful person making a good faith geopolitical argument, instinctively understood that the technology and the person holding it cannot be separated. You cannot evaluate the China parity question without also evaluating who in America gets to operate these systems, under what oversight, with what accountability, and what happens when that person decides the threat is domestic.

The answer to the commenter’s rhetorical question is that no, most Americans would not want these tools in Hegseth’s hands. And that instinct is not a soapbox. It is the correct analysis, stated plainly. An autonomous weapons system and a mass surveillance apparatus in the hands of a Defense Secretary who publicly called Anthropic’s refusal a betrayal, who labeled safety restrictions as woke ideology, and who designated an American company with the same national security label we use for Chinese adversaries, is not a tool for defeating Beijing. It is a tool for something else entirely.

The oldest check on authoritarian power has always been the possibility that human beings in the security apparatus will refuse to follow orders when those orders turn against their own people. History is full of moments where that refusal mattered. Autonomous systems engineered out of existence. You do not need soldiers who might disobey. You do not need surveillance operators who might leak what they see. You do not need anyone to pull a trigger and live with what they did.

Anthropic drew a line around exactly those two capabilities. The United States government designated them a national security risk for drawing it. And the most thoughtful person defending the Pentagon’s position in that forum inadvertently explained why the line matters, in a sentence he described as a different soapbox entirely.

It is not a different soapbox. It is the only soapbox that counts.

And here is the part that virtually no one is talking about, perhaps because it requires stepping back far enough to see the full picture.

Anthropic did not lose a contract. They were banned from all government work, designated with the same national security label we reserve for foreign adversaries, and every company that does business with the Pentagon was told it must certify no contact with Anthropic anywhere in its operations. That is not a business dispute. That is a public execution designed to be watched. Every American technology company with government ambitions saw exactly what happened this week, and they understood the message with perfect clarity: maintain safety standards and we will destroy you, or remove them and we will reward you. OpenAI announced a Pentagon deal for classified networks within hours of Anthropic being banned. The race to the bottom did not begin gradually. It began that afternoon.

Think about what this means for American technology on the global stage. The United States has long told the world that its technology sector is trustworthy precisely because it operates within a framework of law, ethics, and independent corporate standards. American cloud infrastructure, American AI systems, and American software dominate global markets in part because the world believes they are not simply instruments of state power. That belief, worth hundreds of billions of dollars in economic advantage and immeasurable in geopolitical influence, just took a serious hit. Every foreign government, every international enterprise, every allied nation that has been debating whether to build its critical infrastructure on American AI just watched the American government blacklist its most safety-conscious AI company for refusing to build surveillance and autonomous weapons without restriction. The message they received is that American AI companies are ultimately instruments of whatever administration holds power, and their safety commitments last exactly as long as the government tolerates them.

We are told this is about defeating China. But China could not have designed a more effective strategy for undermining American technological credibility than the one the Pentagon executed this week on its own. They did not need to. We did it ourselves, in public, before lunch, and then wondered why no one is applauding.

What I found is not a reason to be concerned. It is a reason to understand exactly what kind of place you are living in, because the answer varies enormously depending on your building, your elevation, and how close you are to the water. This coast sits atop one of Europe’s most significant seismic zones, and the preparedness infrastructure around it is real but unfinished. Knowing the difference between those two things could save your life.

The Eurasian and African tectonic plates are pressing toward each other at roughly four to five millimeters per year southwest of Portugal, roughly the rate your fingernails grow. Over centuries those millimeters accumulate into enormous stored energy, and the release of that energy is why this stretch of Atlantic coast has one of the worst histories of earthquake destruction anywhere in Europe.

There are two distinct threats here, and they are not the same.

The first is the offshore megaquake. The earthquake of November 1, 1755 was most likely generated by a massive fault structure in the Horseshoe Abyssal Plain, several hundred kilometers southwest of Portugal. It had a magnitude of 7.7 or greater and killed somewhere between 15,000 and 60,000 people depending on which scholarly estimate you accept. It destroyed most of Lisbon’s buildings and sent a tsunami across the Atlantic that was still damaging harbors in Ireland and the Caribbean. Based on what we now know about the fault structures involved, Verisk AIR’s seismic modeling group estimates the return period for a repeat of that offshore event at between 3,000 and 4,000 years. That is a long interval. The offshore megaquake is not the most statistically pressing concern for people living on this coast today.

The second threat carries a shorter fuse. The Lower Tagus Valley fault system runs directly through the Lisbon metropolitan area and has generated its own destructive earthquakes, including events in 1344, 1531, and 1909. Verisk AIR found that the return period of magnitude 6 to 7 earthquakes along this fault could be as short as 150 to 200 years. The last significant event was in 1909. We are now 117 years past it, which puts us roughly in the middle of the lower bound of that return period range. We are not overdue. A return period is an average across geological time, not a schedule. What the numbers do tell us is that another serious earthquake on this fault is a realistic possibility within a normal human lifetime.

To understand what that means in practice, consider what the 1755 event actually did. A University of Porto seismic risk framework paper notes that in 1755, more than 50 percent of Lisbon’s building stock was heavily damaged or destroyed, and approximately 10 percent of the population perished. That happened with a city of roughly 200,000 people living almost entirely in unreinforced masonry. The building stock today is different, the population is larger, and the emergency systems are better. But a significant fraction of the older building stock remains, and that fraction is where the risk concentrates.

There are two ways this coast kills people, and they threaten different people differently.

In a major onshore or near shore earthquake, the shaking itself is what kills you. Buildings collapse, and the older the masonry construction, the worse it performs. Distance from the water is irrelevant here. What matters is your building.

In an offshore megaquake generating a tsunami, the shaking may be significant but the deadly part arrives later. In 1755, the first tsunami wave reached Lisbon approximately 40 to 60 minutes after the earthquake. A similar event today would give roughly the same window. The system kills people who do not recognize that the sea pulling back means a wall of water is coming, and people who are in low lying buildings at sea level when the waves arrive. In the Cascais and Estoril area, the coast faces the open Atlantic more directly than central Lisbon. Historical impact research shows the fatality rate in Cascais in 1755 approached 29 percent of its population. Along the Lisbon waterfront, the first tsunami surge pushed roughly 250 meters inland and reached an estimated run up of 12 meters. Contemporary accounts describe water rushing up the Tagus so fast that people on horseback near the docks were forced to gallop toward high ground.

Now I want to talk about buildings, because this is where your personal risk is most determined.

The single greatest factor in whether you survive a large earthquake on this coast is not where you live relative to the water. It is what you are living inside, and the different building generations here perform completely differently under shaking.

The oldest surviving structures predate the 1755 earthquake. These are stone masonry constructions with no seismic design intent whatsoever. They are rare in residential use now but exist in historic cores. In a strong earthquake they are the most dangerous category on this coast.

The Pombalino buildings, constructed after 1755, were built with what was then sophisticated seismic engineering. An internal wooden bracing system called the gaiola pombalina was designed to flex rather than collapse, combining the flexibility of timber with the fire resistance of masonry. These buildings perform better than their age suggests. The critical caveat, documented in research published in Structural Rehabilitation of Old Buildings, is that many have been modified over the centuries in ways that compromise the original design, with alterations to interior walls, addition of heavier floors, and removal of structural elements that were integral to how the gaiola system worked. A Pombalino building that looks intact from the street may have had its seismic resistance substantially degraded by previous owners.

The buildings constructed roughly from the mid 1800s through the 1920s, known as Gaioleiro construction, abandoned the earthquake resistant features of the Pombalino style in favor of taller and cheaper masonry. These are generally the most vulnerable residential structures on this coast. Research published in the Bulletin of Earthquake Engineering found that for this building type, under the standard seismic design scenario for Lisbon, there is approximately a 50 percent probability of heavy damage and approximately a 30 percent probability of collapse. If you live in an unreinforced masonry building from this era, that number deserves your full attention.

The buildings from roughly 1930 to 1960, sometimes called Placa construction, use masonry with reinforced concrete floor slabs. An improvement over the Gaioleiro generation, but still significantly more vulnerable than anything built after 1983. Post 1983 reinforced concrete buildings were built under Portugal’s first serious earthquake engineering code. Post 2010 buildings were designed under European Eurocode 8, which was fully enforced in Portugal since 2022. A properly constructed building from this era is substantially safer than anything older.

Now here is where it gets complicated, and I wonder how many people on this coast are making a mistake.

Walking through Estoril and Cascais, you see older buildings that look completely new. Fresh facades, renovated interiors, new windows, new kitchens. Many of these were marketed as renovated or rehabilitated properties. The question that almost nobody asks when buying one is what exactly was renovated. A cosmetic renovation, even a thorough and expensive one, does not change the seismic performance of the underlying structure. New tiles, new plumbing, new electrical, a repainted exterior: none of that strengthens a Gaioleiro frame. The building underneath the renovation is the same building it was before.

Portugal does have a legal mechanism under Eurocode 8 Part 3, the assessment and retrofitting framework, that can require seismic evaluation and strengthening when rehabilitation works cross certain thresholds of scale, cost, or structural intervention. The Portugal Post’s guide for foreign residents notes that permit checks are becoming stricter, particularly for pre 1950 buildings. But the threshold for triggering mandatory seismic assessment is not always reached in a typical renovation. A property developer who strips a building to its shell, installs new everything, and sells it as renovated may have done something genuinely beautiful without touching the structural system at all.

If you own or are considering buying a renovated older building on this coast, the questions that matter are specific. Was a structural engineering assessment conducted as part of the renovation? Were any seismic strengthening works performed on the load bearing walls, floors, or connections? Is there documentation from a licensed structural engineer confirming the works? If the answer to any of those is no or unknown, you are living in the same building that was there before, with a better kitchen.

For tsunami risk, elevation is your primary protection, and the terrain along this coast creates very sharp gradients. A 1755 scale event would generate waves reaching roughly 6 to 12 meters of run up along the Lisbon and Estoril coast based on historical records and modern modeling.

At sea level within 200 to 300 meters of the water, you are in the primary inundation zone. This includes beach level establishments, the Estoril boardwalk, the Cascais marina, and any property within a short sprint of the waterline. After a major earthquake, you have roughly 30 to 60 minutes before the first wave arrives. Whether you survive depends on whether you recognize the signal and move uphill immediately.

At 10 to 20 meters elevation, 300 to 500 meters from the water, you are in a partial buffer zone. A 12 meter run up can reach further inland than most people expect, particularly through streets or stream valleys that channel the water.

Above 30 meters of elevation, or on high ground inland, tsunami risk drops substantially. In Estoril, much of the residential area climbs the hillside above the coastal train line. If you live above that line at meaningful elevation, your tsunami exposure is low. I live in this range. The water is not what concerns me. My building is. For additional detail you can see this 2015 report that goes though multiple scenarios:

In the hills above Cascais, along the Sintra road corridor (up near N6-8), or anywhere inland of the coastal ridge, the tsunami is essentially not your problem. Your concern is the earthquake itself and what your building does.

Portugal has invested meaningfully in warning infrastructure. Cascais advanced its application for UNESCO Tsunami Ready certification in late 2025, making it one of the first municipalities in Portugal to pursue this designation. The system includes coastal sirens, digital panels, evacuation signage, and integration with IPMA, Portugal’s meteorological and seismic institute, which is an accredited tsunami service provider for the Northeast Atlantic.

Warning systems save lives when people know what to do with the warning. If you feel strong prolonged shaking while on or near the coast, that is your first alarm regardless of whether sirens have sounded. You move uphill immediately. The sea pulling back from the beach is not a curiosity. It is a run signal. Multiple waves follow the first, and the second is often larger. After evacuating, you do not return to low ground until authorities issue a formal all clear.

How does this coast compare to other high risk cities? Verisk AIR’s analysis and Lisbon’s own ReSist program materials place the city in a middle tier globally, clearly below Tokyo, Santiago, Wellington, and Taipei, which have completed or nearly completed citywide retrofit programs and have populations that have internalized earthquake preparedness culturally. But clearly above cities like Istanbul where tens of thousands of dangerous buildings remain unaddressed and political obstacles to retrofit are severe. The closest analogy is San Francisco before the major push following the 1989 Loma Prieta earthquake, combined with more functional early warning infrastructure for the tsunami threat specifically.

The city of Lisbon’s ReSist program is real, involving building level vulnerability assessment, educational campaigns, and prioritized intervention planning. But it is in progress rather than finished, which means the building next to yours may or may not have been evaluated yet.

Here is how I think about my own situation. I know my building, when it was built, what its structural system is, and roughly how it is likely to perform. I know my elevation and I know the route uphill from the waterfront. I have walked it. I know that the sea pulling back is a run signal. I know the indoor earthquake drill: drop, cover, hold until the shaking stops, because running outside during shaking puts you under falling facades. I keep water and basic supplies for several days. Here is an article on how to prepare.

Portugal’s leading seismologists are direct about this: the next large earthquake on this coast is not a question of if. The fault structures are loaded. The plates are moving. What differs between 1755 and the next event is not whether it happens but how many people die when it does. That number depends on building quality, public knowledge, warning infrastructure, and individual behavior, and all four are improvable.

This coast is worth living on. The hills above Sintra rise behind it, the Atlantic opens in front of it, and the light in the evenings does something to the air that I have not found anywhere else. The Portuguese people are amazing. Understanding the seismic reality beneath it does not diminish that. It just adds the layer of knowledge that anyone choosing to live in a place this beautiful and this geologically active ought to have.