What are the US and China actually negotiating about AI?

The reported talks concern crisis communication and warnings about serious AI-related national-security threats.

Officials are discussing a channel for warning each other about AI-related national-security threats ahead of a Trump–Xi summit. The direction is closer to crisis communication than a shared technology program: reduce surprise, clarify when an incident is accidental, and create a way to raise urgent concerns before misinterpretation escalates. The details and enforcement remain unsettled.

Safeguards matter because powerful automated systems can spread errors or trigger decisions faster than normal diplomatic communication can correct them.

Why rivals still need a phone line

Crisis channels between rivals are an old idea: a direct line lets each side ask “was that you?” before suspicion hardens into response. AI makes that more valuable, not less. Automated systems can glitch, misread, or act faster than diplomats can talk, and an accident can look exactly like an attack. A warning channel does not require trust or friendship — it simply lowers the chance of a costly misunderstanding.

What to do, step by step

1. Keep the scope straight: this is crisis communication, not a shared technology program.
2. Separate what has actually been agreed from what is still only being discussed.
3. Watch for a named, specific mechanism rather than warm general statements.
4. Read each government's own account of the talks and compare the wording.
5. Judge progress by whether the channel exists, gets tested, and gets used.