The United States and China plan to hold talks over artificial intelligence in September, five people familiar with the matter told Reuters, as the two powers confront how to manage the risks posed by each other’s increasingly capable frontier models.
The discussions, described as a significant outcome of the Trump-Xi summit in Beijing in May, would likely take place before Chinese President Xi Jinping’s planned September 24 visit to the US, though the dates are not finalized. On the US side the talks would be led by Treasury Secretary Scott Bessent, according to four of the sources, while the agenda, location and other participants remain under deliberation as preparations are at an early stage. The White House, State Department and Treasury did not immediately comment, nor did several Chinese ministries and regulators.
The talks would mark the first official US-China dialogue on AI under President Trump. There is a modest precedent: in 2024, the Biden administration and Xi agreed that humans, not AI, should control decisions on nuclear weapons following working-level talks in Geneva.
The September meeting would go further into the governance of advanced models themselves. Both governments are weighing how to regulate systems that could strengthen military capabilities, enable cyberattacks on critical infrastructure, and disrupt labor markets. Bessent said in May the goal was to halt the proliferation of powerful AI models to non-state actors, and analysts expect the first session to be foundational rather than conclusive.
Anthropic’s Mythos model sits at the center of the concerns driving the talks. The powerful cybersecurity-focused model, currently released only to select US institutions, has alarmed officials in Washington over its potential to find and exploit software vulnerabilities, prompting the Treasury to work with major banks on patching weaknesses.
That worry cuts both ways: Beijing’s priorities for the talks reportedly include Mythos, how the US will control the release of future frontier models, and whether Washington will restrict US companies’ use of Chinese open-weight models, something Axios has reported the administration is considering. China, meanwhile, is mulling curbs on overseas access to its own most powerful models, partly in response to the same dynamics.
Why Both Sides Showed Up
The talks reflect a rare alignment of interests amid intense rivalry. Bessent has framed the US willingness to engage as a function of strength, arguing in May that the two sides can have “wholesome discussions” precisely because the US is in the lead and would not be talking if China were far ahead. For Beijing, the appeal is access to technical understanding of how Washington intends to govern frontier releases, and a forum to press its concerns about export controls and model access.
The shared thread is fear of loss of control, particularly the prospect of criminal or terrorist groups exploiting frontier cyber capabilities against financial systems or infrastructure, a risk neither government can contain alone. One source said China is seeking genuine technical engagement rather than political theater.
Low Expectations, High Stakes
Analysts caution against expecting breakthroughs. As Paul Triolo of DGA-Albright Stonebridge Group put it, a first meeting is unlikely to solve major problems and may at best agree on basic terms, such as a shared definition of a frontier AI model. That modest framing is itself meaningful, since even agreeing on common vocabulary would be a foundation for future cooperation.
The backdrop is fraught: the US has restricted exports of advanced Nvidia chips to China while selectively clearing some sales, the administration is still negotiating voluntary safety standards with US AI companies without having announced binding rules, and the recent episode in which Washington briefly restricted Anthropic’s Fable and Mythos models drew warnings from allies about relying on US technology that could be switched off.
Whether a bilateral dialogue can produce durable guardrails between strategic competitors, or becomes another venue for managing distrust, is the open question September will begin to answer.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
