The U.S.-China 'AI prisoner's dilemma': Losing supremacy may be scarier than AI itself

Lee Kwang-hyung

The author is a former president of KAIST.

Calls to slow the pace of AI development are gaining momentum. The future risks AI could pose deserve attention. Yet two motives may be intertwined behind calls for restraint. One is genuine concern for humanity’s future. The other may be an attempt by market leaders to protect their positions. Regulation that raises barriers before latecomers catch up naturally favors incumbents.

A U.S. president faces a different calculation. Companies may prioritize profits, but a president must consider national survival. That helps explain why the Donald Trump administration has eased AI regulation while accelerating development and adoption. The White House says the United States must “win the AI race,” treating AI leadership as central to competitiveness and security.

The president therefore faces two tasks: containing AI’s long-term risks to humanity and preserving America’s immediate advantage over China. The problem is that these goals can conflict.

AI differs from earlier technologies. It does not remain confined to one industry but cuts across industry, the economy, science and defense. It will shape productivity, military power and research capacity. Falling behind in AI could therefore mean more than losing one industry. It could mean losing global primacy itself.

China has already nearly caught up with the United States. Stanford University’s 2026 AI Index says the performance gap between the best American and Chinese models has effectively disappeared. Their frontier models have traded the lead, and as of March 2026, the top U.S. model led by only 2.7 percent. The United States leads in top-tier models and private investment, while China is strong in research papers and patent filings. If Washington slows while Beijing races ahead, the balance could quickly reverse.

This is where the prisoner’s dilemma emerges. Both countries presumably recognize that if AI becomes uncontrollable, both could suffer catastrophic consequences. Cooperation on safeguards is rational. But neither side can trust the other. If one steps on the brake while the other accelerates, the cautious side falls behind. In an atmosphere of mutual distrust, neither wants to brake first.

For Washington, humanity matters, but China’s pursuit is an immediate strategic challenge. As the Korean saying goes, “The law is far away, but the fist is close.” U.S.-China competition is the urgent problem at hand. This helps explain the Trump administration’s emphasis on AI leadership rather than restrictions. Washington has also made adoption of advanced AI for national security a priority.

Does that mean humanity’s future must remain neglected as long as great-power rivalry continues? Paradoxically, there may be room for hope. China has another priority as important as catching the United States: preserving its system and social stability. If advanced AI were to hack state institutions or cause social disorder through disinformation, Beijing would have strong reasons to impose controls.

Signs are already visible. China recently released its “AI Safety Governance Framework 3.0,” emphasizing risk management and AI that is safe and controllable. This is not solely about regime preservation; user protection and public safety are also involved. The key point is that China, too, is treating AI safety as a national risk.

That creates an opening for an international AI safety regime. History offers precedent. The United States and Soviet Union initially raced to develop nuclear weapons. Once it became clear that those weapons could destroy both sides, however, space for cooperation emerged. The International Atomic Energy Agency was established in 1957, followed by the Nuclear Non-Proliferation Treaty. An international order for managing nuclear risk arose alongside fierce strategic competition.

AI risks may ultimately follow a similar path. The world should establish an international AI organization to share minimum safety standards and verify the development and use of dangerous AI. Agreed red lines are especially necessary for nuclear command systems, biological and chemical weapons management and autonomous AI.

Method matters. Moral appeals for Washington and Beijing to stop competing will not work. Each must recognize that controlling AI risks serves its own national interest. International cooperation does not run on goodwill alone. It advances when a catastrophic threat becomes a shared interest.

Some will ask whether that point will come too late. It is a valid concern. Even as strategic competition continues, U.S.-China talks on safety rules should begin now. Humanity has built imperfect controls for nuclear and biological weapons. There is no reason to assume AI is beyond such efforts. If major powers’ interests converge, law and power can sit at the same table. Managing AI as seriously as nuclear weapons, pathogens and narcotics would itself be a success.

What, then, should Korea do? Its position differs fundamentally from those of the United States and China. Korea still has far to go. It should prepare nimbly for emerging global rules and safeguards, but now is not the time to brake. Korea should press the accelerator and move faster toward technological leadership.