US and China agree an AI incident channel and talks by November

The United States and China will hold regular talks on the risks of advanced AI and have agreed to open a channel for reporting AI incidents to each other, according to a White House fact sheet issued at the end of Chinese President Xi Jinping’s state visit to Washington. The fact sheet, dated September 25, says the next exchange of the new “U.S.-China Super Intelligence (SI) Dialogue” will take place by November 2026.

The same document records a stranger agreement. The two leaders “agreed to use the term ‘super intelligence’ rather than ‘artificial intelligence’” to describe the technology. China’s Foreign Ministry said in a separate statement that Beijing valued Washington’s use of the new term, Reuters reported.

A few lines in a trade document

Most of the fact sheet is about tariffs, coal and fentanyl. Under a new Board of Trade, the two sides reached consensus on better tariff treatment for $30 billion of non-sensitive goods in each direction. The AI passage is short: the naming agreement, a dialogue “to exchange views on risks and benefits related to SI”, a deadline for the next exchange, and a “bilateral communication channel for SI incidents”.

Beijing’s account was similar in spirit. As the technology advances, the Chinese ministry said, the two sides should step up exchanges and work toward consensus “in line with new developments”.

What is missing matters as much. The fact sheet does not mention chips, semiconductors or export controls, the issues that have defined the AI contest between the two countries. It does not say which agencies will run the channel. And it does not say what an “SI incident” is. Axios, which reported the deal, noted that it is unclear which incidents would qualify or what kind of notice either side would give.

Why the new name is more than a quirk

The wording comes from President Donald Trump. In his speech to the UN General Assembly on September 22, he said US documents would switch to “super intelligence” because the word “artificial” makes the technology “sound fake”.

So why should the label matter? In the research world, superintelligence usually refers to AI that is far more capable than people across most tasks. A dialogue named after that idea can drift in two directions: towards the chatbots and agents that are causing real incidents now, or towards hypothetical future systems, which are easier to talk about in general terms. The likelier reading is that Washington wanted its own vocabulary in the text and Beijing saw little cost in accepting it. China’s statement welcoming the term suggests it was glad to hand Trump that point.

A hotline without a definition

Treasury Secretary Scott Bessent focused on creating the channel during preparatory meetings with Chinese officials the weekend before the summit, according to Axios, which said some have likened it to the Cold War-era “red telephone”.

The comparison flatters it. The Moscow–Washington hotline, set up in 1963 after the Cuban missile crisis, existed for one danger both sides understood: nuclear war by mistake. An AI incident has no shared definition yet. It is not obvious whether an American lab’s agent wandering into Chinese government systems would count, or a model that helps a hacking group, or stolen model weights. Until the two sides write that down, each decides for itself when to pick up the phone.

Channels with Beijing, fewer rules in Geneva

A separate report from the same days points the other way. At talks in Geneva in early September on what would be the first agreement governing AI weapons, US and Russian negotiators stripped out language requiring such systems to operate in a “predictable” and “reliable” manner, a clause on ethical considerations, and a provision that humans review AI-generated military targets before a strike, The Washington Post reported, citing three people familiar with the negotiations. The State Department, Russia’s Foreign Ministry and the UN did not return the Post’s requests for comment.

Put the two stories side by side and a pattern appears, though this is our reading and Washington has not commented on the Post’s account. The administration is willing to build a crisis line with its main rival, where each side controls what it reports. It is much less keen on shared rules that would limit how its own military uses AI. Xi, for his part, said during the visit that the two countries have “the capability and responsibility to develop and manage AI for good”.

What to watch

The first SI exchange is due by November. In the same month, governments return to Geneva to discuss whether the weapons talks can turn into an official mandate or a treaty. Trump and Xi are due to meet again at the APEC summit in Shenzhen in November and at the G20 in Miami in December.

The real test comes the first time an incident traces back to one side’s own lab, and that side has to decide whether to make the call.