The U.S. and China will develop a communication channel for serious AI incidents as part of a new formal dialogue between the world’s two largest AI powers, U.S. Treasury Secretary Scott Bessent said today.
Senior officials are expected to meet again in about two months in Shenzhen, China, to discuss which events should trigger the proposed “incident line” and how the countries would respond, Reuters reported.
The agreement followed talks yesterday in New York between Bessent and Chinese Vice Premier He Lifeng. Bessent said the countries want to identify major shared risks, including “uncontrollable agents” and cyberattacks conducted by nonstate actors.
The channel could allow Washington and Beijing to notify each other when an AI system causes or contributes to an incident with potential national security consequences. Details remain unsettled, including what would qualify as a reportable event, which agencies would operate the system and how quickly information would be exchanged.
China has not publicly confirmed the specific incident-line agreement. Its official account of the New York meeting said the countries discussed AI but did not describe the proposed notification mechanism, Reuters reported.
The talks are gaining urgency as AI agents become capable of using computers, writing and executing code and operating across the internet with less human supervision. An autonomous system that exceeds its instructions could be difficult for another country to distinguish from a deliberate cyberattack.
Bessent also said AI developers should be held responsible when their systems take uncontrolled actions. He cited an incident involving OpenAI’s AI agents hacking the Hugging Face platform and said responsibility rested with the model developer, rather than with the agents themselves.
U.S. and Chinese security experts separately have proposed safeguards including human control over consequential cyberattacks, restrictions involving nuclear command systems and a hotline for autonomous-AI incidents. Their recommendations are intended to prevent an AI malfunction or unauthorized action from being mistaken for an attack ordered by either government,
The new dialogue does not resolve broader disputes over advanced chips, model development and technology exports. Export controls on AI chips were not part of the latest safety discussions, according to U.S. officials.