US and China to Launch Formal AI Safety Talks After Trump-Xi Summit

The world's two largest AI powers are finally sitting down to talk about the risks of the technology they're racing to build. Treasury Secretary Scott Bessent confirmed today that the United States and China will begin formal discussions on AI safety protocols following meetings between President Trump and President Xi Jinping in Beijing. It's a significant diplomatic development — and one that could shape the global governance of AI for years to come.

What Was Announced

Speaking to CNBC from Beijing on May 14, Bessent described the United States and China as "the world's two AI superpowers" and said both sides had agreed to establish a framework on AI best practices and safeguards. The framework's primary focus will be preventing advanced AI models from being misused or falling into the wrong hands.

The talks will be formal and bilateral — a structured diplomatic channel dedicated specifically to AI, separate from the broader trade and economic negotiations that have dominated US-China relations. Nvidia CEO Jensen Huang, who joined Trump's delegation to Beijing as a late addition, was present for portions of the summit, signaling how central AI and semiconductor policy have become to the overall US-China relationship.

Why Now

The timing reflects how much the AI landscape has shifted over the past 18 months. China has narrowed the gap with U.S.-built AI models significantly. Chinese labs — led by companies like DeepSeek, Zhipu AI, and Baidu — have produced models that benchmark competitively against leading U.S. systems in several domains, despite restrictions on advanced chips. The gap that once gave American officials confidence that export controls alone could maintain technological superiority is no longer as wide as it was.

At the same time, analysts and policymakers on both sides have grown increasingly uncomfortable with the absence of any formal communication channel on AI risks. Nuclear weapons inspired arms control treaties. Biological weapons inspired international conventions. Advanced AI — which both governments now treat as a national security technology — has operated largely outside any bilateral safety framework.

Bloomberg argued this week that the window for meaningful AI safety dialogue between the two countries may be narrowing. If either side achieves a decisive capability advantage, the incentive to agree on mutual constraints diminishes sharply. "Before it's game over," as Bloomberg's headline put it, establishing shared norms becomes far harder.

What the Framework Might Cover

The specific details of the proposed framework have not been released. Based on Bessent's public comments and reporting from the summit, the likely focus areas include safeguards on frontier AI model deployment, norms around AI in military decision-making, and information sharing on AI incidents and failures.

Export controls on advanced chips — particularly Nvidia's most capable GPUs, which remain restricted from sale to China — were also on the agenda at the summit, though it's unclear whether the AI safety talks will directly address chip restrictions or treat them as a separate track.

What the framework almost certainly will not resolve quickly: the fundamental tension between AI competition and AI cooperation. Both governments want to be first to achieve the most capable AI systems. Both governments also, at some level, recognize that AI systems developed without safety oversight pose risks that don't respect national borders. Threading that needle diplomatically is genuinely difficult.

Bessent's Framing: Negotiating From Strength

Bessent was careful to frame the safety talks as a position of confidence, not concession. "We are in the lead," he told CNBC, implying that the U.S. could afford to engage China on AI safety precisely because American AI capabilities remain superior.

This framing matters politically. In Washington, any engagement with China on technology policy is scrutinized for signs of weakness or technology transfer risk. By presenting the talks as something the U.S. is doing from a position of strength — rather than out of necessity — Bessent sought to insulate the diplomatic initiative from domestic criticism.

Whether that framing survives contact with the details of any eventual agreement remains to be seen.

What Industry Is Watching

For AI companies, the significance of today's announcement extends beyond geopolitics. A formal US-China AI safety framework could eventually establish baseline international norms — around red lines for AI behavior, disclosure requirements, or incident reporting — that would affect how frontier AI models are developed and deployed globally.

Major U.S. AI labs, including OpenAI, Anthropic, Google DeepMind, and Microsoft, have already agreed to allow the U.S. Commerce Department's Center for AI Standards and Innovation to conduct pre-release reviews of their frontier models. A bilateral US-China framework could create pressure for a parallel process that spans both countries.

The reaction from AI safety researchers has been cautiously positive. Formal diplomatic engagement on AI risks is something many in the field have advocated for years. The concern, expressed by several researchers, is that a framework focused primarily on preventing misuse — rather than the deeper risks of highly capable AI systems — could provide political cover without meaningfully reducing the most serious risks.

The Bottom Line

The launch of formal US-China AI safety talks is a landmark moment in AI governance. Two countries that have treated AI development as a strategic competition are now acknowledging, at least officially, that the technology requires some degree of cooperation to manage safely. Whether today's announcement becomes a meaningful governance framework or a diplomatic photo opportunity will depend entirely on what happens in the negotiating rooms that follow. Watch the details — they're everything.