The United States and China have agreed to establish a formal intergovernmental dialogue on artificial intelligence safety. Announced by China’s Ministry of Finance, this marks a significant diplomatic breakthrough following the recent summit between President Donald Trump and President Xi Jinping. The agreement ends a two-year freeze on bilateral engagement regarding AI risks and signals a mutual recognition that the two nations, however fierce their technological competition, share an interest in preventing the most catastrophic potential outcomes of advanced AI development.
China’s Ministry of Foreign Affairs formally confirmed the agreement on Tuesday, stating that the two nations will launch discussions focused on the safe development of AI. U.S. Treasury Secretary Scott Bessent highlighted the importance of the agreement, calling it one of the three “most important achievements” of the Trump-Xi meeting held in Beijing from May 13 to 15. The dialogue is expected to focus heavily on preventing the proliferation of powerful AI models to non-state actors, a shared concern for both Washington and Beijing that transcends their broader geopolitical rivalry.
Ending a Two-Year Silence
The resumption of talks comes at a critical juncture in the global AI race. The last formal dialogue between the two nations on AI safety occurred in May 2024. Since then, the capabilities of large language models and agentic AI systems have advanced at a pace that has surprised even the most optimistic forecasters. The potential risks associated with the deployment of these systems — from autonomous cyberattacks to the development of biological weapons, have grown commensurately, increasing the urgency of establishing shared norms and guardrails.
According to analysis from the AI Safety China Substack, the talks could begin within four to eight weeks. Chinese experts from institutions such as Tsinghua University’s Center for International Security and Strategy (CISS), Fudan University’s China Center for Global AI Governance (CGAIG), and the China Institutes of Contemporary International Relations (CICIR) are expected to play a significant role in shaping Beijing’s approach to the dialogue. The focus on non-proliferation suggests that the initial discussions may center on establishing common standards for model security and export controls, rather than broader ethical or regulatory frameworks.
The Limits of Cooperation
While the agreement to talk is a positive step, significant challenges remain. The U.S. continues to restrict Chinese access to advanced semiconductors, and both nations are actively promoting their respective AI ecosystems globally. As EastFrontier reported in April, the U.S. State Department has been actively warning allies about Chinese AI distillation campaigns, suggesting that the competitive and cooperative dimensions of the U.S.-China AI relationship will continue to coexist in an uneasy tension.
The dialogue will test whether Washington and Beijing can compartmentalize their strategic rivalry sufficiently to cooperate on mitigating the existential risks posed by advanced artificial intelligence. The precedent of nuclear arms control, where the U.S. and Soviet Union maintained dialogue even at the height of the Cold War, suggests that such compartmentalization is possible. But the speed of AI development, the dual-use nature of the technology, and the deep mutual distrust between the two nations make this a far more complex challenge than managing a finite stockpile of nuclear weapons.
What success would look like in the near term is not a comprehensive treaty, but rather the establishment of shared vocabulary, communication channels, and a mutual commitment to transparency on the most dangerous potential applications of AI. Experts at Tsinghua’s CISS have suggested that the initial talks focus narrowly on preventing AI from being used to assist in the development of weapons of mass destruction, a goal that both nations can endorse without compromising their broader competitive strategies. If the dialogue can produce even a limited agreement on this narrow set of issues, it would represent a meaningful step toward the kind of AI governance framework that the global community urgently needs. The stakes could not be higher: as both nations race to develop increasingly capable AI systems, the window for establishing guardrails before the technology outpaces humanity’s ability to govern it is closing rapidly. The agreement to talk, however limited in scope, is a recognition by both Washington and Beijing that some problems are too consequential to leave entirely to competition, and that the first step toward governance is simply the willingness to sit across the table from one’s most formidable rival and begin.