Before Xi's Washington visit, the US pitches China a standing AI-incident hotline -- built to leave chip export controls untouched

Two days after eight hours of trade talks in New York produced no breakthrough on tariffs, US and Chinese negotiators did agree on something new: a proposal for a standing channel to talk about AI. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng concluded the Sept. 20 session at JPMorgan Chase's Manhattan headquarters with a US pitch for what officials are calling the USA-China AI Dialogues -- a notification mechanism letting each government flag AI incidents that rise to a national-security threat. It's the concrete deliverable heading into Thursday's summit between President Trump and Chinese President Xi Jinping in Washington, the two leaders' first meeting since May, when they agreed only to hold higher-level AI talks eventually.

Bessent's framing was procedural, not confrontational. “We think that, just like with any cross-border activity, that moving from opaque to more transparency between the number one and the number two AI powers in the world is very important,” he told reporters, calling the engagement ‘very successful.’ He described the goal more specifically elsewhere: the two sides want to open a communications line, an incident line, where both sides can agree on what the leading AI dangers actually are. China's chief trade negotiator, Li Chenggang, said the talks happened “in a good atmosphere,” with both sides planning to continue.

The dialogue's scope is deliberately narrow. US Trade Representative Jamieson Greer confirmed that export controls on advanced AI chips -- the single most contentious US-China tech issue on the table -- sit outside the new mechanism entirely; that fight stays in its own lane. The follow-up session is billed to cover uncontrolled autonomous agents, non-state actors' misuse of AI, cyberattacks and bioweapons risk -- shared-threat territory where Washington and Beijing have some actual incentive to cooperate, unlike market access or chip sales.

Not every voice heading into Thursday wants a softer table. Six days before Xi's arrival, Rep. John Moolenaar, chair of the House Select Committee on China, wrote directly to Trump urging the opposite approach: strengthening controls on advanced AI products, cutting off cloud services loopholes that allow PRC companies to easily dodge these restrictions, and, in his words, most importantly, stopping exports of American and allied semiconductor manufacturing technology. His letter asked Trump to ensure no one in the administration grants China “any concession or pause in policy action” that could help it catch up in AI -- naming the same summit the AI-dialogue proposal is headed into.

The administration is negotiating from a position where its own public stance treats AI-safety alarm as overblown. Trump has called warnings about AI risk a ‘hoax’ on Truth Social and used the same week's announcement of an 'AI Force' and a still-unfilled AI czar role to signal the administration isn't slowing anything down. That combination -- dismissive of AI risk as a domestic talking point, but willing to build a bilateral incident-notification channel with a rival power -- is less contradictory than it looks: an incident line is a tool for managing an adversary's AI, not a commitment to pace America's own.

“The immediate goal is less about negotiating a major AI agreement... and more about establishing a durable channel for discussing frontier AI risk.” — Alison Szalwinski, The Asia Group

That lower bar matches the pattern of the relationship's other AI friction this month. This newsroom has already reported how a blacklisted Chinese server maker kept buying Nvidia's most advanced chips through a suspended enforcement gap -- precisely the kind of loophole Moolenaar's letter names, still open as Thursday's summit approaches. May's Beijing summit was, by most accounts, heavy on ceremony and light on substance; this one arrives with an actual proposal on the table, even if that proposal is scoped to avoid the fight that matters most to both sides' hawks.

China's own domestic framing is running in parallel, not in response. Xi has told a Shanghai conference that authorities must keep AI under human control, and separately offered BRICS nations help developing open-source AI -- positioning Beijing as a technology partner to the developing world rather than a safety risk. China's Minister of State Security, Chen Yixin, has warned instead about deepfakes and “intelligent bot farms” used to “manufacture political rumors,” a domestic-security framing of AI risk that has nothing to do with Washington. (The only real precedent for a codified US-China understanding on an AI-adjacent capability is the 2024 Biden-Xi agreement that humans, not machines, must control nuclear-weapons launch decisions -- later written into the 2025 National Defense Authorization Act. That took a full year from handshake to statute.)

None of that means Thursday produces a treaty. A notification line is a promise to talk before the next incident, not a limit on what either side builds. But it is the first concrete AI-specific deliverable either government has put forward ahead of a leader-level meeting -- and whether it survives contact with Moolenaar's export-control push, or gets quietly narrowed further, is the actual story to watch once the cameras leave Thursday's room.