U.S. Treasury Secretary Scott Bessent proposes a bilateral U.S.-China AI-incident notification mechanism to increase transparency on AI-related safety events.

US Treasury Secretary Scott Bessent has stated plainly what Washington has long suspected: China would not voluntarily disclose significant AI incidents. The comment, made during discussions with Chinese Vice Premier He Lifeng in New York on September 20, 2026, identifies the core problem that a proposed bilateral AI notification mechanism aims to address.

The talks explored a system where the US and China would notify each other about national security-level AI incidents—creating a formalized channel for the world's two largest AI developers to flag serious events to one another. Bessent framed the goal as building a "shared vision for common goals and threats" between the two countries on artificial intelligence.

This builds on earlier discussions between President Trump and President Xi in May 2026, where establishing a broader "US-China AI dialogue" was first proposed. One significant caveat: export controls on advanced AI chips are explicitly excluded from the proposed dialogue.

Bessent's push for transparency arrives against escalating US concerns about Chinese AI development practices. A central worry involves AI model "distillation," where Chinese firms allegedly leverage outputs from advanced US AI systems to accelerate training of their own models. Bessent himself warned in mid-2026 that China's AI advancements represent America's "biggest risk" in technology.

The exclusion of chip export controls from the dialogue creates a complex dynamic for semiconductor companies. They remain caught between enormous demand from AI development and tightening government restrictions on who they can sell to. By keeping chip policy separate from the AI safety conversation, Washington signals that it views export controls as leverage it does not intend to negotiate away at the safety table.