US Treasury Secretary Scott Bessent said he plans to propose an emergency notification process with China for cases in which artificial intelligence systems go wrong. He told The Axios Show that he believes Beijing would accept a channel designed to communicate rapidly about dangerous incidents.
The proposal follows AI-safety discussions between Bessent and Chinese Vice Premier He Lifeng before President Xi Jinping's state visit in September. Those talks included the idea of an incident-communication channel, but the public comments describe a plan Washington intends to advance, not a completed bilateral agreement.
Bessent said the conversations covered uncontrolled AI agents and the possibility that non-state actors could use advanced models for cyberattacks or biological threats. These risks would affect both countries even as they compete over computing power, frontier models and technological influence.
He argued that China may have underestimated the reach of its own open-source models. As an example, he cited an incident in which the Chinese Kimi model allegedly sent sensitive Chinese data to Anthropic, saying the episode showed Beijing how powerful and difficult to contain the technology can become.
Bessent described leading Chinese models as perhaps eighty or ninety percent as capable as top US systems while being available without comparable guardrails for adaptation anywhere. That combination of broad access and high performance, he suggested, creates shared security concerns beyond the bilateral technology race.
China also worries that AI could undermine the Communist Party's control of information. Its intelligence chief warned that the technology could enable information and cognitive warfare against the government, turning model governance into a question of political stability as well as technical safety.
The United States faces its own tension between rapid development and precaution. Bessent called the administration's approach safe acceleration and said model reviews remain voluntary, though the government reserves the right to intervene if a laboratory proceeds despite serious safety concerns. Responsibility, he said, rests first with the labs.
Asked about executives who fear losing control of powerful systems, Bessent said they should slow down. He urged more attention to resilience and protection of critical infrastructure rather than focusing only on the race to the frontier. A notification channel would test whether strategic rivals can cooperate around narrowly defined common dangers.
