Story
September 22, 2026
Washington Wants an AI Red Phone With Beijing—Even as the Race Accelerates
The U.S. has proposed an AI incident-alert channel with China before the Trump-Xi summit, arguing that transparency is essential when national security is at stake. Beijing has acknowledged AI talks but not endorsed the plan, while the wider safety debate collides with a fierce technology race.
On Sunday in New York, Treasury Secretary Scott Bessent met Chinese Vice Premier He Lifeng with a high-stakes proposal: a notification mechanism for AI incidents that rise to the level of national security. The U.S. also wants a formal U.S.-China AI dialogue, framing the idea as a way to build a “shared vision of common goals and common threats.”1
Bessent’s case was blunt. Between the world’s two leading AI powers, he said, “moving from opaque to more transparency” matters just as it does in any other cross-border activity.2 The concept resembles a modern crisis hotline: not a deal to slow AI development, but a channel to prevent misunderstanding when advanced systems create a serious security event.
China’s public account of the meeting was more restrained. Xinhua confirmed that AI was discussed and called the exchanges “candid, in-depth and constructive,” but did not mention the U.S. alert proposal.3 That silence leaves the central question unanswered ahead of Thursday’s Washington summit between President Donald Trump and President Xi Jinping: whether Beijing will turn a discussion into a standing mechanism.
The talks unfold amid competition, not détente. Washington’s controls on advanced chips and chipmaking equipment remain outside the AI dialogue, according to U.S. Trade Representative Jamieson Greer, while trade negotiators separately explore a category of non-sensitive goods that might receive different treatment in future disputes.4
The safety argument is also running into Trump’s preference for speed. He has resisted calls to curb AI growth on the grounds that restraint could let China catch up, even while announcing plans for an “AI Force” and an AI czar. Safety advocates, meanwhile, argue that rapidly improving systems and an international race make communication indispensable; Carnegie researchers have recommended a continuously staffed, technically capable crisis channel with clear triggers.5
For now, the proposed red phone is less a breakthrough than a test: can two governments racing for AI advantage agree that some failures are too dangerous to discover in silence?