U.S. and China Discuss AI Incident Alerts Ahead of Summit

Bessent Says US and China Discussed AI Alerts; Experts Question Who Would Assess Them

A warning about a dangerous AI system would be hard to act on if the officials receiving it could not judge the risk. That is the concern now facing a proposed US-China alert channel: Treasury Secretary Scott Bessent says the countries discussed setting one up, while people familiar with the plan say it has no technical assessors.

Bessent described a way for either country to notify the other if its AI systems posed a threat or behaved unpredictably. He said the channel could also let the governments discuss common goals. It remains a proposal, not an operating safety system: China had not publicly acknowledged it when Ars Technica published its account.

People familiar with Bessent’s plan told Politico that no officials with technical expertise were assigned to assess the urgency of alerts. Sam Bresnick, a researcher at Georgetown University’s Center for Security and Emerging Technology, told Ars that using an existing channel is better than building a new one. But technical experts on both sides could have a more useful conversation about how they evaluate risks, he said, particularly military ones.

Even with specialists involved, the governments would need to agree on what deserves an alert. Omdia analyst Su Lian Jye said an effective mechanism would require shared categories of risk, triggers for sending notifications, procedures for responding and regular reviews. Those choices would determine whether an alert prompts a useful exchange or simply arrives with no agreed next step.

Su called the proposal a useful start for dialogue, but cautioned that more transparency would not erase the countries’ differences. His point sets a practical limit on the idea: a notification mechanism can provide a place to talk without settling how Washington and Beijing interpret the same event.

The channel would also depend on each government being willing to share information with a competitor. Trump has increased export controls intended to limit China’s access to technology for advancing AI, and Congress is considering further restrictions on chips and hardware. Trump has publicly framed AI leadership as a contest the US must win.

Bresnick warned that China may see US-led safety talks as a way to preserve an American advantage rather than manage a shared danger. Alvin Wang Graylin, a technology strategist and fellow at Stanford’s Digital Economy Lab, likewise argues that treating AI primarily as a race can obstruct cooperation. These are the experts’ assessments of how Beijing might view the proposal, not a stated Chinese response to this channel.

Graylin argues that both countries have reasons to exchange information about risks, including AI-assisted attacks by hackers or terrorists on infrastructure. Bresnick sees value in learning how each government assesses danger, even if he doubts the proposed arrangement will make a major difference on its own. That leaves the immediate test narrower than a broad safety pact: whether the two sides can put qualified people and agreed procedures behind an alert.