A Nobel economist’s new paper on automation and repression raises a disturbing question about what happens when humans become less economically necessary.
1. The uncomfortable question
A new paper, Automation and Repression, published in June 2026 by Nobel Prize–winning economist Daron Acemoglu and his coauthors A. Arda Gitmez and Mehdi Shadmehr, explores a disturbing possibility: what happens to democracy when automation makes human labor less economically important while concentrating more wealth and power in the hands of capital owners?
The paper does not claim that AI will destroy democracy. It builds a mathematical model showing how automation, inequality, political unrest and repression can interact. Under certain conditions, the model allows a democratic society to reach a point where powerful capital owners have an incentive to abandon democracy and establish a more repressive political system. That is not a prediction. It is a mechanism.
And as AI moves toward increasingly capable automation—and potentially AGI—the mechanism raises a much larger question: what happens when the people who own the productive machinery no longer economically need the people who vote?
2. A warning from 1958
Long before anyone was talking about AGI, Aldous Huxley was already worrying about the political consequences of technological progress. Writing in Brave New World Revisited in 1958, he argued that democracy becomes difficult to sustain when economic and political power increasingly concentrates—and specifically connected that concentration to technological progress.
“Democracy can hardly be expected to flourish…”
Aldous Huxley, Brave New World Revisited (1958)
Huxley was not predicting artificial intelligence, but the underlying question is remarkably contemporary. Technology does not distribute power automatically. Sometimes it does the opposite: machines become more capable while ownership becomes narrower, and if productive power becomes concentrated, political power may eventually follow. That is precisely the territory into which Acemoglu’s 2026 paper now ventures.
3. The old bargain between capital and labor
Modern economies have always contained a tension between those who own productive assets—businesses, factories, machines, technology, financial capital and increasingly computational infrastructure—and those whose main economic asset is their ability to work. The distinction matters because economic power is not distributed equally simply because political rights are. A worker and a major capital owner may each have one vote while possessing radically different amounts of economic power.
Historically, however, workers have had something that gave them leverage: they were economically necessary. Factories needed workers, offices needed employees, businesses needed skilled professionals, and governments needed functioning populations and taxpayers. That dependence created bargaining power through strikes, organization, wages and political pressure. Automation changes part of that equation because it can change not only how production happens, but who is necessary for production.
4. When machines start replacing bargaining power
Automation does not merely change how goods and services are produced. If a machine replaces a worker, the immediate effect may look like a productivity improvement; if this happens across an economy, however, the share of output going to human labor can decline while the share going to owners of capital rises. Imagine a company that becomes enormously productive while requiring very few employees. It can generate tremendous wealth, but the workers who once possessed bargaining power because production depended on them no longer have the same leverage.
That changes the central question. The issue is no longer simply “Will AI take jobs?” It becomes: what happens to political bargaining power when human labor itself becomes increasingly replaceable?
5. Inequality becomes a political problem
Acemoglu, Gitmez and Shadmehr’s model begins with a basic conflict. Automation can increase the returns to capital, allowing owners to capture a larger share of the economic surplus as machines become more productive and workers less essential. That does not automatically produce authoritarianism; a democracy has another response available—redistribution. Governments can tax some of the gains from capital and transfer resources to the broader population.
But redistribution creates a conflict between groups with different interests. Workers may want more of it while capital owners may want less, and the more unequal the economy becomes, the more consequential that conflict can be. This is where the paper becomes particularly uncomfortable: its central result is that automation and repression can become complements. As capital becomes more valuable and powerful, the incentive can grow for those who control it to prevent political forces from taking too much of it away.
6. The democracy problem
Democracy complicates the equation because workers are not merely workers; they are also voters. If workers constitute a majority, they can potentially use democratic institutions to redistribute some of the wealth generated by capital. From the perspective of capital owners, that creates a dilemma: accept redistribution, attempt to influence democratic politics, or, under sufficiently extreme conditions, prefer a political system in which workers have less ability to challenge their economic position.
The paper explicitly models the last possibility. Its theoretical framework allows a society that begins as a democracy to reach a point where capitalists have an incentive to overthrow democracy and establish a repressive capitalist state. That sounds like science fiction, but it is a mathematical possibility generated by the assumptions of the model. The distinction is crucial: a mathematical possibility is not the same thing as a forecast.
7. When people stop believing resistance can change anything
There is another piece of this puzzle that the mathematical model cannot fully capture. A population does not have to be physically forced into submission to become politically powerless. It can become apathetic, exhausted, distracted, divided—or simply convinced that nothing will change. Powerful institutions, political actors and media organizations do not necessarily have to suppress every protest; sometimes the information environment itself can overwhelm people with competing controversies, polarization and cynicism about whether collective action can accomplish anything.
The result can resemble learned helplessness at a societal level. People may know something is wrong and even be furious about it, while no longer believing that organizing, protesting or demanding accountability will produce a meaningful consequence. A major scam breaks, a war dominates the headlines, a political scandal explodes, documents appear to implicate powerful people, the internet erupts—and then another controversy arrives. This does not prove that elites or media organizations are consciously coordinating to manufacture apathy. The more interesting observation is simpler: outrage and political action are not the same thing. A population can be furious and powerless at the same time.
“Never have so many been manipulated…”
Aldous Huxley, Brave New World Revisited (1958)
8. From repression to resignation
This creates an interesting extension of the paper’s logic. The model asks what happens when elites face the threat of revolt and discover that repression can protect their position. But imagine the threat of revolt itself becoming weaker because people are economically insecure, politically polarized or permanently distracted. They may fight one another rather than organize against concentrated power, or become so accustomed to corruption, broken promises and institutional failure that each new scandal produces less sustained outrage than the last.
Eventually, the extraordinary becomes ordinary. The population does not necessarily need to be told that it is powerless; it learns the lesson through repeated experience. It protests and little changes, votes and sees limited change, demands accountability and watches powerful people survive. For an exhausted individual, the rational response can become Why bother? A government that must violently suppress a population is facing a powerful source of resistance. A government whose population has largely concluded that resistance is pointless faces a much easier problem.
The most effective repression may eventually be the kind that convinces people they are free to resist—and simply makes them believe resistance is pointless.
“If liberty means anything at all…”
George Orwell, The Freedom of the Press, proposed preface to Animal Farm (1945)
That is the other side of political freedom. Democracy is not merely the right to vote every few years; it also requires the ability to question power, expose uncomfortable facts, organize against institutions and tell the powerful what they do not want to hear. If people gradually lose that confidence—or if the information environment becomes so fragmented that collective action becomes almost impossible—the formal machinery of democracy can remain intact while its substance quietly erodes.
9. The “cuckold” state of politics
There is an uglier psychological version of this. A society can repeatedly watch powerful institutions ignore demands, violate expectations or escape consequences and gradually normalize it. The citizen knows what is happening, is angry and complains, but ultimately accepts that someone else has the real power. That is political resignation: awareness without agency.
It may be more dangerous than ignorance. An ignorant population does not know what is happening; a resigned population knows and has stopped believing it can do anything about it. The critical danger is therefore not the absence of outrage but the normalization of powerlessness.
“love their servitude…”
Aldous Huxley, Brave New World Revisited (1958)
Huxley’s point was darker than simple censorship. A sophisticated system of control might shape incentives, desires, distractions and habits so effectively that people cease to experience their condition as something worth rebelling against. The prison does not necessarily need walls; sometimes the most effective prison is the belief that there is nowhere else to go.
10. A lesson from Orwell
Orwell approached the same problem from another direction. In Animal Farm, power does not have to announce that equality has disappeared; it can change the meaning of equality itself.
“All animals are equal, but some animals are more equal than others.”
George Orwell, Animal Farm (1945)
The brilliance of the line is that the original language survives while its meaning is destroyed. Everyone remains “equal,” the rules remain and the slogans remain, but actual power has become radically unequal. That is relevant to an AI economy in which everyone may formally remain a citizen while a tiny group controls an enormous share of the productive infrastructure. Political equality can survive on paper while economic power makes that equality increasingly hollow.
11. Why AI makes the thought experiment more interesting
The reason this old economic question suddenly feels futuristic is that AI may push automation into areas that previously resisted it. Traditional automation was exceptionally good at replacing repetitive physical or procedural tasks. AI can increasingly perform work involving language, analysis, coding, customer service, research, design and decision support.
If increasingly capable AI eventually performs a large fraction of economically valuable cognitive work, the economic role of human labor could change substantially. That does not mean humans become useless: new industries can emerge, new tasks can be created and humans can become more productive. But there is another possibility: the owners of increasingly powerful AI systems may capture a disproportionate share of the gains.
AI also operates beyond factories and offices. It can be used for surveillance, prediction, identification and information control. The paper itself points toward this possibility: if technology makes repression cheaper or more effective, the political incentive identified by the model becomes stronger.
12. The really frightening question: who needs whom?
For much of industrial history, capital needed labor. Factories required people, businesses needed employees and governments needed workers and taxpayers. That dependence placed a natural limit on how far powerful groups could push society. AI raises the possibility of weakening some of those dependencies by allowing a relatively small amount of capital, computation and automated systems to produce enormous amounts of economic output.
The traditional bargaining relationship then begins to change. The question is no longer simply whether workers can negotiate a higher wage. It becomes: what leverage remains when production can continue without them? Democracy gives people political equality, but that equality becomes harder to defend if economic power becomes radically concentrated and the majority loses the ability to impose meaningful costs on the minority. A society can therefore become technologically richer while becoming politically more unequal.
13. So where does the “one in three” come from?
There is an important distinction here. The paper does not calculate a one-in-three probability of democratic collapse. It provides a theoretical model, not a real-world probability estimate.
But if we step beyond the paper and ask for a rough analytical judgment about the specific mechanism, I would not treat the possibility as negligible. A reasonable way to frame the concern is:
Roughly one chance in three that, in at least some societies where AI-driven automation produces sufficiently concentrated economic power and severe political conflict, powerful elites could successfully move toward repression rather than accept substantial redistribution of wealth and political power.
That is not a one-in-three chance that AI destroys democracy, nor a one-in-three chance that the world becomes authoritarian. It is a deliberately rough estimate of whether the elite-repression mechanism could actually succeed in some societies if the conditions described by the model emerge. The number should therefore be treated as a judgment, not a measurement.
14. Why even a one-in-three possibility is disturbing
The consequences are asymmetric. If AI produces enormous productivity gains and societies distribute them broadly, people could work less, goods and services could become cheaper, science and medicine could advance faster, unpleasant work could disappear, and human capabilities could be amplified rather than replaced.
The opposite outcome has a different character. If technological power becomes concentrated, political institutions weaken, and a small group acquires both enormous productive capacity and increasingly sophisticated tools of surveillance and control, reversing that concentration could become extremely difficult. Once a society loses the mechanisms through which ordinary people can bargain with concentrated power, rebuilding those mechanisms may be much harder than preserving them in the first place.
15. What would the warning signs look like?
The paper’s mechanism suggests that we should pay attention to several developments occurring together: human labor becoming less economically necessary; ownership of productive AI and capital becoming increasingly concentrated; productivity gains flowing disproportionately toward capital owners; political conflict over redistribution; and technological systems making surveillance and political control dramatically cheaper.
There is another warning sign worth adding: people becoming increasingly convinced that collective action is futile. None of these developments individually proves that democracy is in danger. Their combination is what matters. The critical variable may not be AI capability alone, but AI capability multiplied by ownership concentration, political weakness and public resignation.
16. The escape route is not to stop technology
There is an important reason not to read this as an argument against AI. Technology itself is not the villain. The same AI systems that could concentrate power could also distribute it: they can make individual workers more capable, lower the cost of expertise, allow small businesses to perform work that once required entire departments, and help people learn, create, diagnose problems and build things that were previously inaccessible.
The difference is largely about who owns the technology, who has access to it and how its gains are distributed. A future in which AI makes every person more capable looks very different from one in which AI makes a small number of owners extraordinarily powerful while making everyone else economically dispensable. The political debate therefore cannot be reduced to job creation versus job destruction. The deeper question is: how does AI redistribute power?
17. The strange paradox of abundance
There is an old assumption that economic abundance naturally creates freedom. It does not necessarily. A society could become unimaginably wealthy while political power becomes concentrated in very few hands. Imagine automated systems producing most of what society needs: that sounds like paradise until the next question is asked—who owns those systems?
If ownership is broad, abundance could create unprecedented freedom. If ownership is concentrated, abundance could instead make the owners of productive infrastructure extraordinarily powerful. The problem is therefore not simply scarcity. It is control over abundance.
18. The dystopian scenario is not really “AI takes over”
The most frightening scenario may be less cinematic than the usual AGI apocalypse. There may be no robot army, machine government or AI dictator. Humans may remain firmly in charge—and that could be precisely the problem.
A small group of humans could control extraordinarily powerful machines, with those machines amplifying their economic and political capabilities. The dystopia would not be machines ruling humans. It would be humans using machines to make other humans increasingly unnecessary—and increasingly powerless. That is a much more familiar form of dystopia, and perhaps a more plausible one.
19. What Acemoglu’s paper is really warning us about
The most important lesson from Automation and Repression is not that democracy is doomed. It is that economic technology and political institutions cannot be treated as separate systems. Change the technology and you can change the distribution of economic power; change that distribution and you can change political incentives; once political incentives change, institutions themselves can change.
That feedback loop is what makes the paper interesting. Automation can create wealth, wealth can create concentration, concentration can create political conflict, political conflict can create incentives for repression, and repression can then protect the concentration that produced the conflict in the first place. Add a population convinced that protest, voting and collective action rarely change anything, and the loop becomes darker: power becomes concentrated, bargaining power weakens, resistance becomes harder, and resignation makes resistance less likely.
20. AGI could break the mechanism—or accelerate it
There is still another possibility. AGI, if it arrives, could make humans vastly more productive rather than simply replacing them. A person with powerful AI could become capable of doing the work of an entire small organization, allowing millions of people to gain capabilities that today belong mainly to large corporations and institutions. Instead of concentrating power, AI could democratize productive capability.
The outcome therefore depends on something technology alone cannot decide: the institutions surrounding it. Ownership rules, tax systems, competition policy, worker participation, social insurance, access to education and computation, and the ability of democratic institutions to adapt may determine whether AGI becomes the greatest expansion of human agency in history—or the greatest concentration of productive power in history.
21. The choice hiding inside the machine
The temptation is to ask, “Will AI destroy democracy?” That question is too simple. A better question is: what happens when productive intelligence becomes abundant, but ownership of that intelligence remains scarce? And an even harder one follows: what happens when the economic system no longer needs most people as workers, while the political system still needs them as citizens?
That is the tension Automation and Repression puts on the table. The paper does not tell us that AGI will create an authoritarian future; it identifies a coherent economic mechanism through which technological progress can, under certain conditions, make repression more attractive to those who control capital. Alongside that mechanism sits another: people do not have to be physically silenced if they gradually stop believing their voices matter.
The future of democracy may therefore depend on more than preventing authoritarian governments. It may depend on preserving human economic usefulness, broad ownership, institutional accountability and the psychological belief that collective action can still change the world.
22. The final question
Acemoglu and his coauthors have not mathematically demonstrated that AI will destroy democracy. They have done something subtler: they have shown how automation can create economic conditions in which repression becomes more attractive to powerful capital owners—and how, under sufficiently extreme conditions, even democracy itself can become something those owners have an incentive to overthrow.
The paper gives us no real-world probability. Our rough one-in-three framing is therefore not a scientific forecast. It is a way of expressing how seriously the mechanism deserves to be taken if AI-driven automation produces extreme concentrations of economic and political power.
Perhaps that is the question we should carry forward into the AGI era: what happens when humans no longer need to work to produce abundance—but some humans no longer need other humans to produce it? And what happens if, at the same time, those who lose economic power also lose the confidence that political resistance can restore it?
That could be the greatest liberation in human history. Or it could create a world in which bargaining power becomes the scarce resource—not intelligence, not labor, and not even wealth. The good news is that this future is not predetermined by the machines.
The challenge is to ensure that as machines become more powerful, human beings become more powerful with them—not merely less necessary to those who own them.
A final Orwellian thought
“If liberty means anything at all it means the right to tell people what they do not want to hear.”
George Orwell, The Freedom of the Press, proposed preface to Animal Farm (1945)
Because perhaps that is what ultimately has to survive. Not merely the right to vote or the right to speak, but the confidence that speaking truth to power is still worth doing—and that power can still be made to listen.