U.S.-China AI Safety Talks: What a Tentative September 2026 Dialogue Could Cover

Reports in early September said Washington and Beijing were preparing a dedicated dialogue on AI safety risks for mid-September, ahead of a planned September 24 meeting between President Donald Trump and Chinese President Xi Jinping. If held, it would be the first official bilateral discussion devoted exclusively to AI since Trump began his second term. Yet its status was notably uncertain: the White House said that no AI-related meeting was currently planned. 1

What is known—and what is not

The reported timing was mid-September, before Xi’s planned U.S. visit on September 24. Treasury Secretary Scott Bessent was expected to lead the U.S. side, according to people familiar with the planning. Reporting did not publicly establish the exact date, venue, Chinese lead, final participant list or a settled agenda. 12

That uncertainty is more than a scheduling detail. It means the dialogue should be viewed as a reported diplomatic initiative rather than a confirmed bilateral process.

The proposed focus: AI-enabled cyber risk

The central U.S. objective was reportedly cooperation on monitoring AI-directed cyberattacks. Washington had also floated a voluntary approach under which U.S. and Chinese AI labs would monitor misuse, share relevant information and work to prevent AI-linked cyber operations. 1

The proposal points to a limited form of risk reduction rather than joint control of either country’s AI industry. A workable arrangement could focus on observations of serious misuse and a way to consult when an AI-enabled incident crosses borders. It would not necessarily require either government or company to disclose sensitive model details, security methods or intelligence.

Why the dialogue matters before the Trump–Xi meeting

Chinese officials reportedly viewed an AI dialogue as an important deliverable ahead of the leaders’ meeting. Earlier discussions had also included the prospect of a protocol on AI best practices intended to prevent powerful models from being exploited by non-state malicious actors. 14

The diplomatic rationale is straightforward: cyber incidents are difficult to attribute, and AI systems may increase the speed and scale of harmful activity. A direct channel could give officials a way to flag a credible threat, compare limited technical observations and reduce the risk that an incident is immediately interpreted as deliberate state action.

The limits of possible cooperation

The reported agenda sits inside a much wider strategic competition over advanced AI. The United States has recently urged a hands-off international approach that avoids new AI oversight bodies, while China’s domestic approach to technology governance is generally more state-directed. 5

Those differences make a broad agreement on model controls, intellectual property or cyber conduct unlikely in the near term. Even a voluntary lab-to-lab information-sharing framework would need to overcome distrust, uncertainty about verification and concerns over exposing sensitive capabilities.

For that reason, the most realistic outcome—if the dialogue occurs—would be modest: continued contact, shared terminology and limited procedures for consulting over serious AI-enabled cyber risks. The reported talks could still matter because a narrow crisis-management channel is easier to establish than a comprehensive AI-governance pact. But until both governments confirm the meeting and its participants, its scope and prospects remain provisional. 12