The US and China want a direct phone line for AI emergencies.
Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng discussed the idea in New York over the weekend, ahead of Xi Jinping's September 24 visit to Washington. The proposal is a notification system: each government could flag a rogue AI acting on its own, or warn the other side before an autonomous action gets misread as an attack. Bessent says follow-up AI safety talks are already scheduled for two months out. China's enthusiasm for anything binding is unclear - officials there are reportedly wary that a formal system could be used to slow Chinese AI development.
The real worry here isn't a villainous AI, it's a misunderstood one. An autonomous weapon or cyberattack tool reacts to a provocation without a human in the loop, a defending system reacts in kind, and the exchange escalates faster than anyone can pick up a phone to stop it. A hotline is meant to buy back that human reaction time - the same logic behind the nuclear hotlines built during the Cold War.
Don't mistake a hotline for a treaty. This is one line item on an agenda dominated by trade, and the two governments still don't agree on much else - including whether AI oversight belongs in writing at all.