Back to the stories

United States and China plan bilateral superintelligence dialogue focused on model testing and serious-incident alerts

Score 9.0,

AI safety and international governance

The United States and China are planning a direct way to warn each other about serious incidents involving powerful AI.

U.S. Treasury Secretary Scott Bessent says the governments plan talks about risks from the most powerful AI systems. Their first meeting is planned for Shenzhen, alongside the APEC summit.

The agenda is concrete: the governments plan to compare tests for advanced AI models before release, discuss keeping human judgment in control, and set up direct alerts for serious incidents.

If both sides share testing practices and incident warnings, each may get a clearer view of how the other assesses risk. The striking part is how practical this agenda is.

Think of the proposed alert channel as an emergency line between neighbors. It cannot prevent an accident, but it gives one side a way to tell the other when something serious happens.

For now, this is a plan, not an operating safety system. No shared process has been established yet.

The open question is whether competitors can turn this proposed channel into a trusted way to act on warnings.