U.S. to propose emergency AI notification system with China

Scott Bessent, the Treasury Secretary, went on "The Axios Show" to unveil the administered version of a fire drill: a "notification process" between Washington and Beijing for when AI goes wrong. Not a treaty, not a moratorium, not a shared kill switch — a mechanism for telling the other side that the thing has already happened. Bessent says he thinks Beijing would agree, which is the geopolitical equivalent of "she'll definitely text back." Note the ambition: a hotline for the apocalypse that doesn't require anyone to stop building the apocalypse.

The risks said to justify this arrangement are uncontrolled AI agents and non-state actors using the technology for cyber or biological threats — the exact roster Microsoft warned about, where the finding was that attackers already have the head start. Now the plan is for two governments to swap notes after the fact. One pictures the ransomware crew filing its notification in triplicate, and whoever releases the engineered pathogen being courteous enough to CC Treasury. Meanwhile the words "binding," "verification," "pause" and "liability" appear nowhere in Bessent's pitch, which is a remarkable feat of vocabulary management. The hotline gets a launch date before the standard does.

Set it beside the rest of the week: the UK government has two months to answer MPs and peers on AI law, and Sen. Mark Warner has filed a bill that would at least create a federal AI safety board. Everyone, everywhere, is building notification architecture for a disaster nobody has agreed to slow down. Still, credit where it's due — this is the first AI accord in living memory that might fit inside a text message. Two superpowers, one group chat, no kill switch, and a mutual promise to describe the wreckage in a timely fashion. Bring a pint; the notification will arrive before the ambulance.