Security2 publishers2 min readPublished
Washington proposed an AI incident notification mechanism to Beijing
Scott Bessent named the proposal on Sunday, four days before Trump meets Xi at the White House. On the record, its one stated boundary is AI incidents that could affect national security.
The Watch · Security desk
What happened
- Treasury Secretary Scott Bessent said Sunday that the United States has proposed a new notification mechanism for artificial intelligence incidents that could affect national security.
- Xinhua said Bessent and Vice Premier He Lifeng talked about issues relating to AI without mentioning specifics, and had candid, in-depth and constructive discussions on economic and trade issues.
- Bessent and Trade Representative Jamieson Greer said the two sides had agreed to operationalize a new Board of Trade that Trump and Xi discussed when they met in Beijing in May.
- Greer said the two countries were still discussing which nonsensitive items could get lower tariffs, and that the list could include consumer and agricultural goods, energy products and medical devices.
Compiled by The WatchSomething wrong?How this is made
Why it matters
- constraint As described, the proposal puts no reporting obligation on any company, so escalation paths and disclosure runbooks stay as written on the strength of Sunday's remarks.
- contradiction Washington named a notification mechanism; the Chinese account referred only to issues relating to AI. The public record does not settle whether this is agreed or merely proposed.
- decision Firms weighing an internal AI incident taxonomy can still set their own thresholds, since they have no intergovernmental definition to conform to or retrofit against.
- precedent Reporting incidents after they happen is the cheapest form of AI cooperation available to a US administration that will not slow development, so it is the form that gets tried first.
A notification channel only matters operationally once four questions have answers: what counts as an incident, at what severity it crosses the reporting line, how fast the notice goes out, and who receives it on the other side. The proposal as described on Sunday answers none of them [15], and what Bessent gave was the intent. "We want a shared vision of common goals and common threats," he told reporters [2]. He said that "moving from opaque to more transparency between the No. 1 and the No. 2 AI powers in the world is very important" [3]. Bloomberg Law reported that he spoke after hours of talks with the vice premier in New York [13].
The verbs attached to the weekend's other outcome were firmer. "Today those notions have become reality," Greer said of the Board of Trade [8]. The AI item was still a proposal in the same set of remarks, delivered in the lobby of JPMorgan Chase's headquarters late Sunday [6].
After-the-fact reporting is also the form of cooperation that costs Washington least. Trump has resisted calls to slow AI development, saying that would help China catch up to U.S. companies [5]. A notice sent once an incident has happened does not interrupt a training run or delay a deployment.
An incident definition is what would connect any of this to internal practice. It would supply a severity threshold to map detections onto, and a clock to build an escalation path around. Both are still missing for the operators of frontier or agentic systems who would eventually sit under such a regime [15].
George Chen, a partner and chair of digital practice at The Asia Group, wrote that "The fact that both sides agreed to continue the dialogue is significant" [11], and that an AI risk notification mechanism "sets a precedent that other countries may follow" [12].
What to watch
- A Chinese readout that names the notification mechanism, instead of referring to issues relating to AI in general.
- Any published text defining what counts as a reportable AI incident, and which agency on each side receives the notice.
- Whether the AI item lands in the same deliverable as the tariff list Greer described, or drifts to a later working group.