The United States and China are preparing to convene, in mid-September, the first official bilateral dialogue devoted exclusively to artificial intelligence since the start of Trump’s second term, according to two people briefed on the planning who spoke to Reuters. Treasury Secretary Scott Bessent is expected to lead the U.S. delegation. A White House official told Reuters that “there is currently no planned AI-related meeting in mid-September,” a denial that itself has the shape of a negotiation in flux.

The timing isn’t incidental. The talks would land in the window before a September 24 summit between President Trump and Xi Jinping in Washington, and days after both governments signed the Carolina Principles at the G20 innovation summit in North Carolina. That document, aimed at discouraging the development of A.I.-specific regulations, amounts to a rare show of tech alignment between the two powers. It commits both capitals to a governance posture that’s, in practice, the absence of one.

Washington’s proposed agenda makes the philosophy explicit. American negotiators intend to argue that U.S. and Chinese laboratories should “police themselves” and share information to prevent A.I.-directed cyberattacks, an extension of the voluntary framework Trump established by executive order in June, under which the labs set their own guardrails. The U.S. side also plans to raise Chinese distillation of proprietary American models, and the prospect that China may soon field a system comparable to Anthropic’s Mythos. The criteria for Washington’s own pre-release cybersecurity reviews haven’t been released.

Two analysts quoted by Reuters read the moment as consequential. “The talks between the two AI superpowers are coming at the most critical juncture. It is now or never,” said Paul Triolo of DGA-Albright Stonebridge Group. Samm Sacks of New America put the technical case bluntly: “We are at a tipping point where frontier agents can cause massive damage when unmonitored. Both the U.S. and China are vulnerable.”

The structural read is straightforward. A governance regime built on lab self-policing doesn’t eliminate compliance risk; it relocates it. When the labs write the rules, the residual exposure lands on the businesses using the outputs, a dynamic already visible in the recent F.T.C. directive on algorithmic pricing disclosure and in the federal court’s rebuke of the Pentagon over its Anthropic contracting. The Carolina Principles ratify a bet that the frontier will be governed by its builders. Mid-September will test whether Washington and Beijing can agree on what that means when the systems misbehave.

Sources