US and China Set September AI Talks, Bessent to Lead US Side

A formal US-China conversation about frontier AI is finally getting a date on the calendar. According to Reuters, citing five people familiar with the matter, the two governments are planning to hold intergovernmental AI talks in September, most likely before Chinese President Xi Jinping's planned September 24 visit to the United States. Four of those sources said Treasury Secretary Scott Bessent would lead the US delegation.

The interesting choice here is Treasury. AI-safety diplomacy could plausibly sit at State, Commerce, or the White House science office, and the fact that it is being fronted by the department that runs sanctions, export-linked financial rules, and the bilateral economic file with Beijing tells you how Washington is framing this. The talks are described as a significant outcome of the Trump-Xi summit in May, and the substantive hook Bessent has publicly given is narrow: a protocol on 'best practices for AI to make sure non-state actors don't get a hold of these models,' as he put it in May, together with the line that the US can afford these discussions 'because we are in the lead.'

That framing matters because it sets the ceiling on what September can realistically produce. The reporting points at a first formal AI channel between the two governments rather than anything resembling arms control. Preventing frontier model weights and capabilities from leaking to criminal or terrorist users is a relatively narrow, mutually convenient goal, and it is also the easiest thing for both sides to agree to talk about without touching export controls, chip rules, or each other's domestic AI industries.

The honest caveat is how thin the confirmed detail still is. Reuters' sources say the agenda, the location, the rest of the participants, and even the exact dates are still being worked out, and the story is sourced to unnamed officials on both sides rather than an on-the-record announcement. What the reporting does not give you is who leads for China, whether the 'non-state actor' framing quietly includes Chinese private labs or only bad-faith misuse, or what any resulting 'protocol' would look like in practice between two governments that do not share evaluations or weights.

Still, if the meeting happens on anything like this timeline, it becomes the first real bilateral surface on frontier AI risk, and everyone from US labs to third-country regulators has reason to watch what language comes out of it, because a US-China baseline will be harder to argue with than any of the parallel frameworks currently in flight.