China’s Biggest Weakness On AI

By Carl Benedikt, Project Syndicate | August 12, 2026

When a state can credibly promise that labor displacement will not mean destitution, workers and their representatives are more likely to accept technological change. But since China can promise no such thing, its lack of a robust welfare state has emerged as a fundamental weakness in the broader AI race.

OXFORD—In late April, a Chinese court ruled that an AI firm could not legally dismiss an employee simply because his role had been automated. Shortly thereafter, Chinese regulators reportedly suspended new autonomous-vehicle licenses nationwide in response to a mass robotaxi failure in Wuhan, a city where cab drivers had long petitioned authorities to slow the technology’s rollout. And all the while, Chinese state media has been criticizing companies for contemplating AI-driven layoffs, arguing that they have a duty to protect their employees.

Together, these episodes suggest that China is beginning to pump the brakes on its own AI rollout. But why would a state with such coercive capacity and a strong desire to dominate the technological frontier choose restraint? After all, China seemed to have a structural advantage in the AI race. An authoritarian state can be a deployment machine, overriding worker resistance, suppressing protests, and keeping the gears turning, whereas liberal democracies are deliberation machines that must account for elections, unions, assembly rights, and other institutional constraints.

But China has a fundamental weakness: it lacks a robust social safety net. True, the West’s high social spending—on unemployment benefits, public health care, and retraining schemes—is usually framed as a brake on innovation, since it requires taxing high earners and transferring wealth from productive sectors to dependent ones. But the historical record shows that social insurance can accelerate the deployment of new technologies. When a state can credibly promise that displacement will not mean destitution, workers and their representatives are more likely to accept technological change.

Of course, plenty of countries have chosen repression instead, clearing the way for the machines by force. Britain deployed the army against the Luddites in 1812 and made machine-breaking a capital offense. The industrializations of South Korea and Taiwan relied partly on suppressing independent unions. But repression is easiest to sustain when it comes with a credible promise of progress. South Korea and Taiwan could suppress labor because they were also delivering factory jobs, urbanization, and rising living standards. Workers were asked to endure low wages today in exchange for richer lives tomorrow.

AI displacement does not allow for such a trade-off in China. It is arriving in offices, services, and platforms, threatening educated young workers as much as factory hands, and it comes just as China’s promise of ever-greater opportunity is becoming harder to keep. Chinese growth is slowing, the property ladder is broken for many young people, and labor-market anxieties are reflected in high precautionary saving rates and nostalgia for the boom years.

Chinese authorities seem to recognize the problem. But even if they understand that technology-driven displacement must be managed politically, China’s lack of a strong institutional cushion limits their options. Public social spending is only around 10% of GDP, less than half the OECD average. Hundreds of millions of internal migrants remain disadvantaged by the hukou system, and Chinese households still bear a large share of health-care costs out of pocket.

As the economist Barry Naughton has shown, China has built much of the framework for a universal welfare system, but transfer flows remain modest, constrained by the urban-rural divide and demographic aging. Its unemployment insurance was designed for cyclical rather than structural displacement, and only a relatively small share of those out of work draw benefits.

China has paid the price for failed technological deployments before. For centuries, craft guilds resisted mechanization to protect their members, thereby delaying China’s industrialization for two centuries. The risk today is similar. The short-term political need to avoid disruption could deprive Chinese society of the latest technology’s long-term benefits.

By contrast, Britain could sustain mechanization during the Industrial Revolution partly because it boasted a social-welfare system unmatched anywhere in Europe. Under the Old Poor Law, parishes, from 1601 onward, were legally obliged to relieve the able-bodied poor. As a result, most paid cash allowances to those out of work or on low earnings, covered rents, supplied bread and fuel, and paid for medical care.

These measures worked. Recent research by the economists Avner Greif and Murat Iyigun finds that English counties with higher per capita poor relief experienced fewer riots as the economy industrialized. The Poor Law had many flaws, but its political consequences were clear: it acted as a shock absorber that helped keep factories running. Far from hampering technological deployment, the welfare state may be an essential complement. Yet the West risks squandering this advantage. In the United States, unemployment insurance is reaching a shrinking share of the unemployed just as AI has begun to affect the labor market.

Of course, some social protections create their own challenges. Many parts of Europe risk delaying AI deployment because there are rules on the books protecting specific jobs rather than insuring workers through transitions. When a German works council demanded that its employer bar staff from using ChatGPT, the Hamburg Labor Court rejected the petition, but the firm still had to litigate the matter. Such uncertainty makes experimentation costly. When SAP restructured its operations around AI in 2024, shedding roughly 8,000 positions, it had to set aside €2.2 billion ($2.5 billion) for voluntary buyouts and early retirement—equivalent, by one estimate, to more than three years of pay per affected worker in Europe.

Denmark’s flexicurity model comes closest to striking the right balance: Weaker employment protections are offset by generous unemployment benefits and intensive active-labor-market policies, including retraining and re-employment. Firms can deploy AI safely, knowing that workers are insured and that the overall political economy will hold.

The race between the US and China is not only about chips, compute, and model parameters but also about institutional capacity. China must now build the welfare state it has long avoided building—and at a time of slowing growth and deteriorating demographics. America and Europe are starting from a better position, but they still have a long way to go to modernize their welfare states and eliminate veto points before displacement becomes politically uncontainable.

Repression is what states reach for when they have nothing else. The AI revolution will be won not by the country most willing to force workers aside, but by the one that makes them secure enough to let the machines in.

Carl Benedikt Frey, Associate Professor of AI & Work at the Oxford Internet Institute and Director of the Future of Work Program at the Oxford Martin School, is the author, most recently, of How Progress Ends: Technology, Innovation, and the Fate of Nations (Princeton University Press, 2025).

Copyright Project Syndicate