U.S. proposes AI incident notification channel with China ahead of Trump-Xi talks

The United States has proposed a new channel for notifying China about serious artificial intelligence incidents that could affect national security, a small but significant step in the effort to build guardrails between the world’s two leading AI powers. AP reported that Treasury Secretary Scott Bessent raised the idea during weekend meetings in New York with Chinese Vice Premier He Lifeng, ahead of planned talks between President Donald Trump and Chinese President Xi Jinping at the White House.

Bessent described the proposal as a “notification mechanism” for AI incidents that could create national security risks. The goal is not a broad technology treaty or a slowdown in AI development. It is a crisis-communication tool: a way for Washington and Beijing to tell each other when an AI-related event crosses a threshold where silence or misinterpretation could make a dangerous situation worse. Bessent said the two countries need a shared view of common goals and common threats, adding that transparency between the world’s top two AI powers is important.

The proposal lands in a tense policy environment. The Trump administration has repeatedly argued that the United States cannot afford to slow AI development because China would continue moving ahead. At the same time, officials on both sides have reasons to worry about accidents, misuse and escalation. AI systems are now being discussed not only as consumer products and business tools, but as technologies with implications for cyber operations, military planning, surveillance, disinformation and critical infrastructure.

A notification mechanism would be modest compared with arms-control treaties, but its importance lies in process. In nuclear and military history, communication channels have often mattered because they reduce the chance that one side misreads a test, accident or warning as hostile intent. AI incidents are harder to define. They could involve a model enabling a cyberattack, a system malfunctioning in a sensitive government context, unauthorized access to a frontier model, or an automated tool behaving in ways that create cross-border risk. Before any mechanism can work, Washington and Beijing would need to agree on what counts as serious enough to report.

That will be difficult. The United States and China define security threats differently, and both have incentives to conceal capabilities. Neither side will want to reveal proprietary model details, intelligence methods or weaknesses in critical systems. The practical challenge is to design a reporting channel narrow enough to protect secrets but specific enough to prevent miscalculation. Even an imperfect system could help officials build habits of communication before a real crisis.

The AI discussion was part of wider economic talks that also covered trade and tariff issues. Bessent and U.S. Trade Representative Jamieson Greer said the countries had agreed to operationalize a new “Board of Trade” first discussed by Trump and Xi earlier this year. Chinese state media described the AI dialogue as candid and constructive, a phrase that signals disagreement but also a willingness to keep talking.

For the AI industry, the proposal shows how quickly AI safety has become international diplomacy. Until recently, many safety debates centered on company policies, model evaluations and domestic regulation. Now the question is whether rival powers can create enough communication to manage incidents without slowing the competition they both see as strategic. A notification mechanism would not solve the AI race, but it could create a first piece of shared infrastructure for managing its risks.