Sunday evening, the lobby of JPMorgan’s New York headquarters. Treasury Secretary Scott Bessent and Trade Representative Jamieson Greer step in front of reporters and run through a day of talks with Chinese Vice Premier He Lifeng. Alongside tariffs and rare earths: a notification channel for AI incidents.
«We want a shared vision of common goals and common threats», Bessent said. His reasoning: like any other cross-border activity, moving from opaque to more transparency between the world’s number one and number two AI powers matters.
What would actually get reported
The proposal sits inside a «U.S.-China AI Dialogue» both sides discussed. The notification mechanism would be its first concrete piece. Bessent limited it to incidents rising «up to a national security level».
That was the whole definition. No threshold, no time window, no named office doing the reporting. China’s state agency Xinhua afterwards confirmed only that the two had covered matters «relating to AI», plus «candid, in-depth and constructive» talks on economics and trade. No mechanism appears in the Chinese version at all.
The rest of the evening
Greer said both sides have now operationalised the «Board of Trade» their leaders discussed in Beijing in May: «Today those notions have become reality.» Which nonsensitive goods might see lower tariffs is still open — consumer goods and low-tech items from the Chinese side, energy products, agricultural goods and possibly medical devices from the American one.
Trump and Xi meet at the White House on Thursday. Trump has no time for an AI slowdown, saying it would let China catch up, and on Friday he called the industry’s safety warnings a hoax. Four days later his own administration offers Beijing a safety channel.
Why it counts anyway
George Chen of The Asia Group, who chairs its digital practice, sees the continued talking as the real result: the fact that both sides agreed to keep the dialogue going is significant, and the first outcome — an AI risk notification mechanism — sets a precedent other countries may follow.
China striking a different note elsewhere doesn’t change that. State television spent last week picking apart Anthropic’s privacy policy. Apparently both things run in parallel.
The threshold is missing
A notification channel without a threshold is a statement of intent with a phone number attached. Who decides that an incident touches national security? Who files the report when filing makes their own lab look bad? That is exactly where the industry’s voluntary rules have gone soft every time so far.
Still: until Sunday there was nothing at all between the two biggest AI powers. A phone nobody answers yet beats no phone. The real test comes the first time a model does something neither side wants to explain.