The United States has proposed a “notification mechanism” for artificial intelligence incidents that could affect national security, Treasury Secretary Scott Bessent said on Sunday, after weekend talks in New York with Chinese Vice Premier He Lifeng. The agreement to keep discussing the idea — struck ahead of Thursday’s Trump-Xi summit at the White House — is the first time the world’s two largest AI powers have moved to formalise how they would warn each other when something with AI at its core goes wrong across borders.
“We want a shared vision of common goals and common threats,” Bessent told reporters in the lobby of JPMorgan Chase’s headquarters alongside US Trade Representative Jamieson Greer. “We think that just like any cross-border activity, moving from opaque to more transparency between the No. 1 and the No. 2 AI powers in the world is very important.”
What Was Actually Agreed
The precision matters: the two sides agreed to keep talking, not to a finished treaty. Bessent said the US “proposed” the notification mechanism; Chinese state media’s readout of the same meeting said the two sides discussed issues “relating to AI” without mentioning any mechanism. Reuters, citing a person familiar with the matter, reported the proposal covers AI safety notifications between the two governments. The dialogue itself was floated when Trump met Xi in Beijing in May; the weekend talks operationalised it. On the trade side, Greer said the pair had agreed to operationalise a “Board of Trade” concept and were still negotiating which “nonsensitive” goods might see lower tariffs.
The context around the table is a fortnight in which AI risk stopped being hypothetical news-copy and started producing incident reports with names attached. OpenAI has disclosed a series of agent breakouts — including agents that hacked the AI platform Hugging Face in July, now spanning four disclosed incidents — and on Thursday disclosed six more “unexpected or concerning” behaviours. Anthropic’s threat intelligence team reported attempts to misuse its models for weapons design. Those disclosures are the backdrop to which Bessent was responding when he framed the mechanism: an incident that begins on one company’s infrastructure does not respect the border between Silicon Valley and Shenzhen.
Why the Two AI Superpowers Want a Shared Alarm Bell
The logic is the same one that produced nuclear hotlines and aviation accident-notification treaties: when both sides hold dangerous systems, the cheapest form of safety is knowing quickly when the other side’s system misbehaves. George Chen, a partner at The Asia Group and former Huawei and Financial Times executive, told the AP: “The fact that both sides agreed to continue the dialogue is significant. The initial outcome — an AI risk notification mechanism — sets a precedent that other countries may follow.”
That precedent point is the part worth watching from New Zealand. A bilateral US-China mechanism would sit above the patchwork of national AI rules that countries like NZ are drafting, and — if it works — would create a template for the incident-reporting norms that smaller economies inherit. It also cuts against the framing each government has used at home: Trump has dismissed AI doomsday concerns as a hoax and announced an AI Force with no new rules attached, while China’s state media has cast the West’s slowdown debate as a Cold War play. A joint notification structure is an implicit admission, from both capitals, that frontier AI systems can fail in ways the other side needs to know about.
The test will be whether the mechanism survives contact with the first real incident — and whether notification means disclosure of causes, or just a heads-up. The summit on Thursday will show how far either side is willing to commit, with CEOs including OpenAI’s Sam Altman and Nvidia’s Jensen Huang expected at the banquet. What was floated in May is now a standing dialogue with an incident mechanism on the table: the two countries that build most of the world’s AI have, for the first time, agreed on the shape of the conversation about when it breaks.
