U.S. and China set up Cold War-style AI hotline

The United States and China have established a Cold War–style hotline to enable direct, rapid communication between senior officials on artificial intelligence risks and incidents. The channel is intended to reduce the chance that accidents, technical failures or misinterpretations of AI behavior escalate into wider diplomatic or security crises.

The secure, dedicated line will be used for rapid notification and consultation when AI systems cause or appear to cause harms that could affect national security or bilateral stability. Scenarios covered include large-scale disinformation campaigns, unexpected behavior in deployed AI systems that affects critical infrastructure, and incidents that might be mistaken for hostile action. Officials expect the hotline to shorten response times and allow immediate clarification before public statements or crisis-level responses are are made.

Negotiations to create the hotline followed months of diplomatic and technical exchanges. The channel will be staffed by designated officials with authority to exchange factual information and to request technical assistance or verification. The mechanism is modeled on historical crisis hotlines from the Cold War and updated to meet modern cybersecurity and encryption requirements. Technical teams on both sides will support the hotline to verify reported incidents and to interpret technical data rapidly.

The agreement sets protocols for authentication, message formats and expected reply timeframes so messages do not get lost in bureaucratic channels. It includes provisions for secure transmission of technical logs and other data needed to evaluate events quickly. Complex technical questions can be escalated to subject-matter experts who can analyze models, data flows and system behavior.

The hotline also provides a channel for exchanging information on emergent threats such as large-scale misuse of generative models or cross-border harms from automated decision systems. Both governments will monitor performance and may refine operating procedures through use and periodic reviews.

The mechanism follows earlier interactions between the two capitals on AI governance, safety testing and norms of behavior. Differences remain on industrial policy, investment and intellectual property, but the two governments agreed on the need to reduce immediate risks that could lead to miscalculation.