AI IncidentsPublic record of AI control failures
Back to published register
Risk signalPolicy warningNot an incident

US proposes an AI incident notification mechanism with China

U.S. Treasury Secretary Scott Bessent said after talks with Chinese Vice Premier He Lifeng that the United States proposed a bilateral notification mechanism for AI incidents that reach the national-security level. An official Chinese readout confirms discussion of AI-related issues, but not acceptance of the proposal or any thresholds or procedures. The mechanism is expected to be put to Donald Trump and Xi Jinping for consideration at their planned summit.

Published
Sep 21, 2026
Signal type
Policy warning
Confidence
98%
Organization
U.S. Department of the Treasury, Government of the People's Republic of China
Sources
3
Last reviewed
Sep 21, 2026

Assessment

What this signal changes

The proposal is a concrete policy signal that the two largest AI powers are treating some AI events as cross-border security matters that may warrant a direct information channel. Bessent specifically described notifications for shared goals and threats when an AI-related incident rises to the national-security level. The Chinese side confirmed that AI-related issues were discussed but did not publicly address the proposed mechanism. The relevant development is therefore a formal U.S. initiative, not an existing bilateral agreement.

Evidence boundary

What the evidence does not show

There is no signed mechanism, joint statement, published definition of a reportable AI incident, or Chinese commitment. The proposal is still expected to be considered at the presidential level. The sources establish a diplomatic initiative and AI talks, not implementation, information sharing, or a new autonomous AI incident. This signal therefore does not change incident counts or the incident timeline.

Classification: This page records a sourced warning, research finding, policy action, or industry response. It is deliberately separate from the Incident register and does not prove that an unauthorized AI action occurred.