The US and China agree to discuss AI incidents without resolving the trade truce

US Treasury secretary Scott Bessent and trade representative Jamieson Greer agreed with Chinese vice-premier He Lifeng to establish an AI-safety and national-security dialogue, likely to meet again within two months. One proposal is a mechanism for notifying the other government about serious AI-related incidents.

That is much narrower than a joint agreement to slow frontier development. Incident notification can reduce the chance that a cyber event, military model failure or unexpected system behaviour is misread as deliberate escalation. It does not require the countries to reveal model weights, accept common capability limits or trust each other’s compliance with a development cap.

The talks occurred inside a more contentious economic negotiation. An extension of the tariff truce remains unresolved; Washington is dissatisfied with Chinese rare-earth commitments and raised alleged Chinese support for Iran. The US also proposed a “board of trade” through which tariffs on non-sensitive goods might be reduced.

The safety channel is therefore best understood as a confidence-building mechanism between strategic competitors, not a comprehensive AI accord. Its value will depend on definitions, verification and whether either side reports incidents that expose meaningful vulnerabilities.