A report published September 21 said U.S. Treasury Secretary Scott Bessent had described discussions between U.S. and Chinese officials about a possible AI dialogue mechanism and notifications for incidents that could affect national security. The proposals would give the two governments a way to communicate about serious risks as they compete over AI.
Bessent described a possible dialogue and incident alerts
The reported discussions covered a channel for AI dialogue and a notification mechanism for incidents with national-security implications. Both are proposals. An incident notification system would let one government alert the other to a serious AI-related event, rather than leave it to infer what happened from incomplete or misleading signals.
Why a military AI incident could escalate
A hypothetical scenario behind the military safeguards is an AI-related intrusion, or an automated defensive response, being mistaken for an attack ordered by the other country. If either government retaliated before understanding the event, a technical incident could turn into a wider crisis.
Misattribution—mistaking who caused an event—raises the stakes. A direct communication channel could give officials a way to clarify what happened before they respond. The scenario is a risk proposal, not a reported attack.
The safeguards proposed for military AI
Melanie W. Sisson and Tianjiao Jiang published a commentary on human control of military AI on September 9. The proposals attributed to their work include setting boundaries around U.S. and Chinese nuclear systems, keeping human control over consequential cyber operations, and creating a dedicated military channel for discussing AI-related incidents.
These proposals focus on crisis management: keeping people involved in high-consequence decisions and opening a line of communication if an AI-related event is misread. The separate notification mechanism discussed by U.S. and Chinese officials also concerns incident communication, but applies to incidents with national-security significance.
AI rivalry leaves room for narrower cooperation
The dialogue proposals sit within a wider competition over AI leadership. Dario Amodei has separately argued for pacing improvements in frontier AI capabilities so safety work can keep up. That proposal addresses the speed of AI development; the military safeguards address human control and communication between governments.
Those are different policy questions. A broad effort to slow AI development would reach much further than sharing information about serious incidents or setting boundaries for military use. The reported discussions focus on the narrower risk-management problem.
