US Proposes AI ‘Incident Hotline’ to China as Rivalry Meets Safety Fears

U.S. Treasury Secretary Scott Bessent used two days of high‑level economic talks in New York to float a new idea to Beijing: an AI “incident notification mechanism” between the United States and China. The proposal, raised with Chinese Vice Premier He Lifeng, would create a channel for both governments to alert each other when artificial intelligence systems behave unpredictably or trigger security concerns.

According to U.S. officials, the mechanism is meant to reduce the risk that a malfunctioning military or government AI system could be misread as hostile intent, echoing Cold War–era hotlines designed to prevent nuclear miscalculation. The plan is expected to be considered by President Donald Trump and President Xi Jinping at their summit later this week.

The move comes as the two countries race to outdo each other in advanced AI chips and frontier‑scale models, even as both publicly acknowledge fears about loss of control and escalation. Still, major questions remain: what kinds of incidents must be reported, how much technical detail would be shared, and how to reconcile transparency with national security. With Chinese officials wary of exposing sensitive systems and some U.S. lawmakers skeptical of any data‑sharing, the talks highlight how hard it will be to turn broad AI‑safety rhetoric into concrete bilateral cooperation.