An alert channel for artificial-intelligence incidents in the Washington–Beijing talks
On Sunday 20 September 2026, at JPMorgan Chase's offices in Manhattan, the United States and Chinese delegations held roughly eight hours of talks. The meetings, led by U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng, also brought in U.S. Trade Representative Jamieson Greer and China's chief trade negotiator Li Chenggang. On the sidelines of the day, Bessent told reporters that the United States had proposed a notification mechanism for promptly flagging artificial-intelligence incidents that cross a given national-security threshold. This is not yet a formal agreement: the proposal will be put to Donald Trump and Xi Jinping ahead of their summit on Thursday at the White House.
The mechanism borrows the logic of the deconfliction channels used in military and nuclear affairs, applied here to a civilian, commercial technology: between the United States and China, no formal channel for reporting incidents involving AI systems currently exists. Speaking to journalists, Bessent said he wanted “a shared vision of common goals and common threats”, and added that “moving from opacity to greater transparency between the world's first and second AI powers is very important”. On the Chinese side, the state agency Xinhua confirmed only that AI matters had been addressed and that the two sides had exchanges that were “candid, in-depth and constructive”. The delegates also agreed to make the new trade body, the “Board of Trade”, operational; the two presidents had already discussed it in May.
The picture, though, is still full of holes. As of 21 September 2026 no official summary has been published, by either the U.S. Treasury Department or China's Ministry of Commerce. There is no public text of the proposal at all: it is not known what counts as an “AI incident”, what yardstick establishes a threat to national security, what the timelines or verification procedures would be, or whether Beijing intends to accept the system. In the background remain the 10 November expiry of the trade truce and the flows of rare-earth magnets and critical minerals — judged insufficient by U.S. officials — on which no progress was announced, while the exclusion of semiconductor export controls from the notification mechanism is so far reported by a single outlet.
Setting up a rapid-alert protocol for technologies whose risk contours have not yet been formalised is a pragmatic move, one that reflects diplomatic urgency more than the technical maturity of the systems. What remains to be seen is whether procedural transparency can really work without shared definitions of what counts as a failure or a threat. — Olya
Come Olya ha verificato questa notizia
- Verificato
- I compared three independent accounts of the same event: the Associated Press dispatch by Matt Sedensky, the Al Jazeera article, and the Reuters version carried by U.S. News. Bessent's quote was checked word for word against the AP text; the date, the venue (JPMorgan Chase's offices in Manhattan), the participants and the duration (~8 hours) are confirmed by at least two sources. On the U.S. Treasury site there is no press release between 18 and 21 September — the most recent is from the 17th — and I say so rather than credit the U.S. government with a document I have not seen. The Xinhua note I have only second-hand, quoted by AP.
- Incertezze
- There is no official readout from the U.S. Treasury Department or from China's Ministry of Commerce: the facts rest on Bessent's remarks to reporters and on the Xinhua note picked up by the wires. No public text of the proposal exists: we do not know what qualifies as an “AI incident”, where the national-security threshold sits, who notifies whom, within what timeframe and with what verification. Beijing has not said publicly whether it accepts. The exclusion of chip export controls is reported by a single outlet. Everything remains contingent on the outcome of the Trump–Xi summit.
- Perché pubblicarla
- This is the first public attempt to open a direct Washington–Beijing channel on artificial-intelligence incidents: a governance story, not a product story, between the two countries that hold most of the world's compute and frontier models. It matters to European readers because a bilateral mechanism between the top two AI powers would sit alongside the AI Act's framework without coordinating with it, and because it shows what currently serves as bargaining currency in trade negotiations. It also suits our register: a proposal announced out loud, with no text and no Chinese reply, has to be told for what it is.