September 17, 2026, (Inside AI) — U.S. Treasury Secretary Scott Bessent said Washington is willing to discuss shared artificial intelligence risks with Beijing during talks scheduled for this weekend, according to a statement obtained by Axios. The remarks, made Wednesday, signal a possible opening in a relationship that has grown increasingly tense over semiconductor controls and military applications of AI.
Bessent is set to meet Chinese Vice Premier He Lifeng this weekend, ahead of a planned meeting between President Donald Trump and Chinese President Xi Jinping next week. The timing places AI governance squarely within a broader diplomatic push.
The Treasury secretary framed the U.S. position as one of strength, not concession. "The United States remains the leader in AI, and we are open to discussions on avoiding shared risks and avoiding bifurcation of our two systems. We expect our discussions to cover both open and closed weight models," Bessent said in the statement.
The reference to open and closed weight models is notable. Open-weight systems allow public access to model parameters, while closed-weight models keep them proprietary. The two countries have taken divergent regulatory paths on this issue. The U.S. has largely allowed open-weight development to flourish, while China has imposed stricter controls on certain model releases. A dialogue on this topic could touch on export controls, licensing, and safety testing protocols.
Bessent acknowledged the broader climate of suspicion. "There's a lot of mutual distrust between the U.S. and China and how each side views the others AI advancements," he said, according to the statement.
That distrust has concrete roots. The U.S. has restricted advanced chip exports to China for years, citing national security concerns. China has responded by accelerating domestic chip production and developing alternatives to American hardware. Both nations have invested heavily in military AI applications, from autonomous systems to intelligence analysis.
The weekend meeting between Bessent and He Lifeng is not the first attempt at AI dialogue. Previous talks have yielded limited results, often stalled by disagreements over verification and enforcement. A senior fellow at a Washington-based think tank, speaking on condition of anonymity, said the challenge is not starting conversations but sustaining them. "Both sides have domestic constituencies that view any cooperation as weakness," the fellow said. Inside AI could not independently verify whether this weekend's talks will produce a formal framework or merely an agreement to continue talking.
The stakes extend beyond bilateral relations. AI governance remains fragmented globally. The European Union has enacted comprehensive AI regulations. The United Kingdom has taken a more sectoral approach. A U.S.-China understanding, even a limited one, could influence how other nations craft their own rules. It could also affect the development of international safety standards, which currently lack enforcement mechanisms.
Bessent's statement did not specify what shared risks the U.S. wants to address. Potential topics include accidental escalation from AI-enabled military systems, misuse of generative models for disinformation, and the security of AI supply chains. Each area presents unique verification challenges. Unlike nuclear arms control, AI development is diffuse, fast-moving, and often conducted by private companies rather than governments.
The meeting between Trump and Xi next week will likely overshadow the Bessent-He Lifeng talks. But the lower-level discussion could set the agenda or establish working groups. Historically, U.S.-China diplomatic breakthroughs often begin with technical exchanges before reaching the presidential level. The AI policy landscape has become a key component of that diplomatic choreography.
Whether this weekend's talks produce anything more than a joint statement remains uncertain. Bessent's language was carefully hedged. He said the U.S. is "open to discussions," not committed to outcomes. The phrase "avoiding bifurcation" suggests a desire to prevent a complete split between American and Chinese AI ecosystems. That split is already underway in some areas, particularly in semiconductor supply chains and cloud infrastructure.
For now, the AI industry watches and waits. A successful dialogue could reduce regulatory uncertainty for companies operating in both markets. A failed one could accelerate decoupling. Either way, the decisions made in the coming weeks will shape the trajectory of global AI development for years.
