US, China to meet again on AI safety in two months in Shenzhen, Bessent says - Reuters

So the planet's two leading AI powers have agreed to meet again in about two months, in Shenzhen, to discuss artificial intelligence dangers, and the person announcing it is Scott Bessent — the Treasury Secretary. Not a science adviser, not a commerce official, not anyone whose brief nominally covers the technology that might one day end the conversation entirely: the bond-yield man. Which tells you roughly where AI policy sits in the filing cabinet — somewhere between tariffs and a sanctions waiver, with 'urgent' scrawled in pencil.

The centrepiece of Bessent's Monday remarks is a formalised dialogue including an 'incident line' for communications about AI safety incidents. An incident line. A phone number you ring *after* the incident. One admires the honesty: it is an admission that incidents are expected, and that the plan is largely to describe them to one another afterwards, in two languages, at speed. Two months is a lovely interval — long enough for a checkpoint to be trained, open-weighted, downloaded and deployed by someone who never got the memo, short enough that no dignitary has to reschedule a holiday. The Soufan Center's proposed US-China cooperation framework we covered at least had the decency to name who was misusing what.

What is being agreed, once you strip the padding, is a protocol for talking about talking. The dialogue is 'formalised' — the great bureaucratic verb, implying that everything preceding it was casual banter over canapés. And the venue for round two is Shenzhen, which is a bit like hosting the fire-safety seminar in the fireworks factory: the place where the stock is manufactured, boxed and occasionally mis-stacked. Bessent offered no date, no delegation list, no agenda and no consequence should the line ring and nobody answer. But the communiqué will be lovely. Probably in two months.