What it is
The AI safety notification mechanism is a proposal, not yet an agreement, for the United States and China to alert each other when an artificial intelligence system or AI-related development poses a risk to national security. It emerged from talks held in New York on September 20, 2026, between US Treasury Secretary Scott Bessent and Trade Representative Jamieson Greer on one side, and Chinese Vice Premier He Lifeng and chief trade negotiator Li Chenggang on the other. The session ran roughly eight hours and, according to Bessent, was intended to prepare groundwork for a Trump-Xi summit in Washington later that week.
As described by US officials, the mechanism would function as a communication channel covering “common goals and common threats” — a way for each government to flag AI incidents that cross a national security threshold, rather than a general-purpose regulatory or safety-testing regime. Several outlets have drawn an explicit comparison to the Cold War-era hotlines established between Washington and Moscow: a dedicated line for signaling danger, not a mechanism for jointly setting rules. US export controls on advanced AI chips and semiconductor manufacturing equipment were explicitly excluded from the talks, meaning the proposal is narrower than the broader technology-competition disputes between the two countries.
As of this writing, no operating rules, activation threshold, or implementation date have been made public. The two governments agreed only to continue discussions — a working group was set to meet again on the Monday following the initial talks — and China’s own public readout of the meeting did not mention the US proposal at all, a discrepancy worth noting in its own right.
Why it matters for AI governance and narratives
The proposal sits at the intersection of two narrative threads the observatory already tracks: great-power AI competition and the search for AI safety governance mechanisms. Its framing is notable for what it is not — it is not a joint safety standard, a testing protocol, or an agreement to slow capability development. It is transparency infrastructure aimed at crisis avoidance between the two dominant AI powers, closer in kind to nuclear-era risk-reduction measures than to the multilateral AI-safety-summit process (Bletchley, Seoul, Paris) that has otherwise defined intergovernmental AI diplomacy since 2023.
That framing matters because it lets both governments claim engagement on AI safety without conceding ground on the issues that actually divide them — chip export controls, competitive access to compute, and each side’s suspicion that the other’s safety rhetoric is a vehicle for constraining its rival’s progress. Bessent’s own language — “moving from opaque to more transparency” — casts the mechanism as a confidence-building measure between rivals who assume the other is developing dangerous capability in secret, rather than as a shared safety commitment. Chinese officials’ silence on the proposal in their readout is itself a data point: it suggests Beijing may not yet be prepared to validate the US framing publicly, even if working-level talks continue.
Key facts and dates
- September 20, 2026: Talks held in New York between Bessent and Greer for the US, and He Lifeng and Li Chenggang for China; the AI notification mechanism was proposed by the US side during this session.
- Prior context: Trump and Xi reportedly first discussed the possibility of AI-related consultations in May 2026 during a Beijing meeting, but that forum was never formalized.
- Scope: The mechanism as proposed would cover AI incidents rising to a national security concern; chip export controls were explicitly kept off the agenda.
- Status: Unresolved. No public agreement, rules, or timeline exist. A working group was expected to continue talks the following Monday, ahead of a Trump-Xi summit planned for later that week in Washington.
- Reception: Experts quoted in coverage were skeptical that a substantive agreement would emerge from the summit itself, citing a “high level of distrust” between the two governments and China’s past wariness that Western calls to slow AI development are aimed at containing Chinese progress specifically.
Where to learn more
- Bessent proposes AI safety notifications in talks with China ahead of Xi-Trump meeting — CNN Business
- U.S. proposes exchanging AI safety alerts with China, Bessent says — NBC News
- US proposes AI safety notification mechanism in talks with China — Al Jazeera
- Bessent proposes U.S.-China AI safety notifications in talks with Chinese vice premier — The Spokesman-Review