The US and China will hold their first direct AI talks in September, and Beijing's agenda reveals what it fears most

US-China AI Talks: Beijing's September Agenda Maps Its Biggest Fears

The United States and China plan to hold their first bilateral AI dialogue in September, and the agenda Beijing is bringing tells you more about the talks than any outcome will. Treasury Secretary Scott Bessent will lead the American delegation, four sources told Reuters 1. The meeting is expected before President Xi Jinping's planned September 24 visit to the United States, though dates have not been finalized 1. It would be the first official US-China dialogue on AI under President Trump 1.

Beijing's priority list, according to analysts, includes three items: Anthropic's Mythos model, how the US will control the release of future frontier models, and whether Washington will restrict Chinese open-weight models 1. Read as a set, these are not neutral diplomatic questions. They map two vulnerabilities central to Beijing's AI strategy: access to the frontier, and the open-weight channel that lets Chinese models reach developers worldwide.

On the frontier side, Chinese authorities are reportedly considering curbs on overseas access to their most powerful domestically developed models, citing concerns about the hacking potential of Anthropic's Mythos, which is not available to the general public 1. Beijing wants to understand how Washington decides which models ship and under what conditions, because that answer determines whether Chinese labs stay within reach of the frontier or get structurally locked behind it.

On the open-weight side, Axios has reported that Washington is considering restrictions on US companies' use of Chinese open-weight models 1. If those restrictions take hold, the open-weight models Chinese labs distribute globally lose their primary access point to Western developers. For Beijing, open weights function as a distribution channel, a way to embed Chinese models into the global developer stack. Severing that channel is a strategic threat, not a regulatory inconvenience.

The irony is that both sides are erecting barriers while asking the other to lower theirs. The US has already restricted exports of advanced AI chips to China 1. China is weighing limits on outbound access to its own models. September is where those counterstrategies collide.

Paul Triolo, a partner at DGA-Albright Stonebridge Group, told Reuters that the initial meeting will likely focus on something deceptively consequential: "This is the first meeting, they're unlikely to solve any major problems. I think they will try to agree to some basic terms, like what defines a frontier AI model" 1.

That definition is the leverage point. Set the threshold high and many capable Chinese models escape any control regime. Set it low and the net widens to catch mid-tier systems. Either way, the definition shapes export controls, investment decisions, and which models are legally available in which markets.

The Trump administration is already in discussions with US AI companies about voluntary safety standards for new model releases, though no regulatory measures have been announced 1. In May, Bessent said future talks would aim to halt the proliferation of powerful AI models to non-state actors 1. Domestic safety standards and bilateral non-proliferation talks converge at whatever frontier definition emerges in September.

One source told Reuters that China wants "real engagement and technical discussions on AI issues from the US rather than political talks" 1. The participants, full agenda, and location are still under deliberation 1.

The priorities Beijing has already surfaced tell you what September is really about. China wants to shape the rules before the rules constrain Chinese AI. Whether it succeeds may determine whether open versus closed stays a business choice or hardens into a geopolitical fault line.