Beijing confirms the first US–China AI dialogue — held inside the trade talks, with nothing published

The confirmable core is one sentence. At a regular briefing on 24 September, Ministry of Commerce spokesperson He Yadong said that under the China–US economic and trade consultation mechanism, the two sides held their first dialogue on AI. That is the whole of the official Chinese characterisation. The ministry did not say when it took place, who sat in it, what was on the agenda, or whether anything was agreed. It placed the dialogue inside the eighth round of trade consultations, which matters: AI was handled as a trade-track subject under trade negotiators, not as an arms-control or safety-regulator channel.

Xi's own words are on the record in full, published by the Chinese government. At the White House arrival ceremony he said both countries are leading nations in artificial intelligence, with both the capability and the responsibility to develop and manage AI for good and to ensure its development is always under human control. That was the only reference to AI in the address. The foreign ministry readout carries the same sentence and records the two leaders agreeing to build a constructive relationship of strategic stability — a formulation that mentions no AI commitment at all.

What is not in any primary document is the part most likely to be repeated as fact. Reporting describes a mechanism for exchanging safety alerts on AI incidents touching national security, a follow-up meeting in Shenzhen in about two months, and an extension of the trade truce to 10 January 2027. None of those appear in a US or Chinese government text. The White House news feed carried no AI item on 24 or 25 September; the entries for those dates concern the First Lady's programme and the state dinner.

The limit is the story. A first dialogue held inside a trade mechanism, confirmed in one sentence by one side's commerce ministry, with no joint text, no named secretariat, no schedule and no scope, is the thinnest possible form of bilateral AI engagement. It establishes that the subject is now formally on the table between the two governments that between them host essentially every frontier-scale training run. It establishes nothing about thresholds, notification triggers, verification or enforcement. Read against the same week's Security Council session, where the United States rejected international AI governance outright, the bilateral channel looks less like a step toward multilateral rules than the alternative to them.

• Confirmed MOFCOM spokesperson He Yadong, 24 September: under the China–US economic and trade consultation mechanism, the two sides held their first dialogue on AI. South China Morning Post
• Confirmed Xi's arrival-ceremony remarks contain a single AI sentence, on developing and managing AI for good and keeping it under human control. State Council of the PRC
• Reported A trade-truce extension to 10 January 2027, a Shenzhen follow-up in about two months, and an AI incident alert exchange appear only in press reporting, in no government text. TheNextWeb