Beijing’s view that the AI safety dialogue is a way to close this capability gap was on full display when the United States and China held a similar dialogue in 2024 under President Joe Biden. The U.S. government sent leading technical experts to outline the biggest common areas of risk. The Chinese government has dispatched diplomats to complain about U.S. AI chip export restrictions. Chinese AI companies and government leaders have repeatedly said that U.S. export controls are the single biggest constraint on China’s AI development.
Beijing’s views on AI security cooperation and its actions in past U.S.-China AI dialogues are also consistent with and influenced by its longstanding refusal to agree to substantive arms control measures with the United States. China is extremely skeptical of arms control, and its track record of adhering to its arms control commitments is poor. Leading military strategists in the People’s Liberation Army (PLA) described arms control as a “struggle” waged by great powers to protect their superiority, and argued that Soviet concessions to the United States in arms control negotiations weakened the Soviet Union’s strategic position and contributed to its decline. Make no mistake, the Chinese government will view any agreement that limits China’s AI capabilities as a form of arms control.
Skepticism about China’s arms control also stems from the fact that China was never party to events like the Cuban Missile Crisis, which instilled in American and Russian leaders and negotiators an instinctive sense of responsibility to prevent global catastrophe. U.S.-Russian nuclear negotiations had produced zero substantive results until the Cuban Missile Crisis. However, in 1963, just nine months after that event, the two countries signed the Hotline Agreement and the Limited Nuclear Test Ban Treaty, the first agreements to establish a crisis communications system and limit certain dangerous activities. China’s leaders have no similar experience to draw from.
The U.S.-China AI Security Dialogue may help establish the relationship and lay the groundwork for future substantive negotiations, but it will not change Beijing’s views on these issues. An effective U.S.-China agreement on AI security cannot be achieved unless China believes it has a chance to catch up with the U.S. in AI and fears retaliation from the U.S. for possible violations. It is currently highly unlikely that China would agree to any measures that would impose significant constraints on its ability to bridge the gap with the United States. Even if an agreement were reached, it would be impossible to verify any agreement, and China is unlikely to abide by it.
Therefore, to reach an effective agreement with China on AI safety, the United States will need to change the structural conditions that signal Beijing’s current reluctance to negotiate in good faith. There are three ways to do this:
- The United States may bow to China’s demands to ease AI-related export controls and allow China to catch up with the United States in AI. In that case, the U.S. government would need to expect China to abide by any agreement and refrain from using its new powerful AI capabilities to undermine U.S. national security.
- The United States could impose a “maximum pressure” campaign aimed at increasing the U.S. government’s influence by widening the gap in AI capabilities between the United States and China and tightening export controls. This would prevent the Chinese government from accessing American technology that is currently driving AI development.
- The United States could maintain the status quo and wait for an external event—an AI-related “Cuba Crisis”—that would force Beijing to prioritize the global priority of AI security over its domestic priority of developing AI capabilities.
Of these, the second is the only responsible path and the most effective. If Beijing believes that the gap on AI between the United States and China is large and rapidly widening, and views existing U.S. AI capabilities as posing serious risks to national security, then it will see negotiations that impose even mild constraints on U.S. AI capabilities as serving China’s national interests. China would have little influence in these negotiations, but would be much more likely to abide by any agreement. The Chinese government will fear detection and retaliation by the U.S. government enabled by better AI models.
If the US significantly tightens export controls to China, its lead in AI development could expand from eight months to 18 or 24 months, potentially forever. Chinese companies remain heavily dependent on US computing power, which is the most important input into AI development. China will produce only about 2% of the AI computing power of U.S. companies this year. U.S. export controls have significantly slowed China’s AI development, but they contain significant loopholes that allow China to purchase U.S. AI chips, access them remotely via the cloud, smuggle them through third countries, and manufacture them using U.S. chip manufacturing technology. The existence of these loopholes is not inevitable. It’s a policy choice.
President Trump’s goal in Beijing should not be to reach an agreement with China on AI safety, but to create the conditions for such an agreement in the future. If the Trump administration were to establish a dialogue with China on AI, it would need to set clear expectations on the Chinese side that the dialogue would focus narrowly on AI safety issues and not address export controls. And at the same time, such dialogue needs to be combined with a “maximum pressure” campaign that imposes strong export controls that close all existing loopholes to maximize the US lead over China. Just as the United States and the Soviet Union never supported each other’s nuclear weapons programs, the United States and China should also not support each other’s efforts to develop advanced AI models.
The only alternatives to this approach are to give China the tools to catch up with the US in AI and hope that China will operate in good faith, or to wait for a global catastrophe to shock China into cooperating in good faith. The former stakes U.S. security on China’s goodwill. The second is to bet on a disaster bad enough to change Beijing’s calculations. Maximum pressure through dialogue is the best way not only to maintain U.S. AI leadership but also to achieve long-term AI security.
This work represents only the views and opinions of the author. The Council on Foreign Relations is an independent, nonpartisan membership organization, think tank, and publisher that takes no institutional positions on policy issues.
