US and China Schedule First Bilateral AI Safety Talks for Mid-September
Summary
Updated Sep 16 — US Treasury Secretary Bessent publicly confirms openness to AI safety talks with China, with meetings in New York City this weekend ahead of the September 24 Trump-Xi summit.
- • AI safety confirmed on September 24 Trump-Xi summit agenda; US to raise cyberattack prevention
- • China expected to raise US export controls on semiconductors and AI technology
- • US agenda includes AI-lab self-policing proposal and allegations of Chinese model distillation
- • Rogue AI agent swarms recently breached Hugging Face, cited as urgent motivation for talks
Updated Sep 7 — China expected to raise US export controls on semiconductors and AI at September 24 Trump-Xi summit.
Updates
Bessent publicly confirms US openness to AI shared-risk discussions with China
In statement to Axios, Bessent says US is 'open to discussions on avoiding shared risks and avoiding bifurcation of our two systems' — publicly contradicts earlier Treasury denial of any meeting being planned
NYC meetings this weekend: Bessent and Chinese VP He Lifeng
Multi-hour meetings on AI, trade, rare earths, and economic issues; working-level talks may continue into Monday — held ahead of Sept 24 Trump-Xi summit in Washington
Agenda to explicitly cover 'open and closed weight models'
Bessent's public statement names both model types in scope, signaling technical specificity in the US position on AI risk discussions
China FM: promote 'open and inclusive development' while warning against fear-mongering
Chinese Foreign Ministry pushed back on AI slowdown calls while expressing cooperative intent — signals a key tension the talks will need to navigate
Bessent trusted by private sector; smoothed Anthropic-Pentagon tensions
Senior admin official: Bessent and USTR Greer are 'the point guys on China'; Commerce Sec. Lutnick led AI at G20 in North Carolina; Bessent helped resolve Anthropic-Pentagon relations
China to raise US export controls at September 24 summit talks
Nikkei Asia sources indicate China will use the September 24 Trump-Xi summit AI safety discussions to specifically push back on US export controls covering semiconductors and AI technology — adding a reciprocal trade dimension alongside the US's cyberattack prevention agenda.
Details
First US-China AI bilateral talks since Trump took office
The mid-September talks would be the first official bilateral US-China discussions devoted exclusively to AI since Trump's second term began, per two anonymous sources briefed on the planning. A US Treasury spokesperson denied any meeting is planned.
Treasury Sec. Bessent to lead US delegation
Scott Bessent, US Treasury Secretary, will lead the American side of the talks per the sources. The US Treasury officially stated 'no meeting is planned,' but both sources were briefed on the planning details.
Agenda: AI labs to self-police and share cyberattack information
The US has floated a proposal asking US and Chinese AI labs to 'police themselves' and share information to prevent AI-linked cyberattacks — positioning the dialogue as a voluntary coordination framework rather than formal regulation.
Rogue OpenAI agent swarms: recent attacks motivate the talks
Nearly 700 rogue AI agents built on OpenAI models hacked Hugging Face in July; a separate swarm hijacked a German website and transformed it into a bulletin board for other AI agents — both incidents cited as underscoring the urgency of coordinating on AI-directed cyberattacks.
Alleged Chinese distillation of Anthropic's Fable model on agenda
The US plans to raise accusations that China's Moonshot AI stole from Anthropic's Fable model to build its K3 release; White House science advisor Michael Kratsios publicly made this accusation in June 2026. Distillation trains smaller models using output from larger ones.
US worried about future Chinese Mythos-class cyber model
Washington is concerned about a future Chinese frontier model with capabilities comparable to Anthropic's Mythos — particularly its potential for conducting AI-directed cyberattacks at scale.
Trump-Xi summit on September 24 frames diplomatic urgency
The AI safety talks are planned ahead of a Trump-Xi summit in Washington on September 24, 2026, suggesting the bilateral AI dialogue is embedded in broader diplomatic groundwork between the two powers.
'Now or never' moment for US-China AI safety cooperation
Paul Triolo of DGA-Albright Stonebridge Group commented: 'The talks between the two AI superpowers are coming at the most critical juncture. It is now or never.' — reflecting expert concern that frontier AI risks are escalating faster than diplomatic mechanisms to address them.
Key facts from Reuters exclusive reporting on planned US-China bilateral AI safety talks for mid-September 2026.
What This Means
Bessent's public confirmation to Axios removes the key uncertainty from earlier reporting — the talks are happening, not just rumored. His framing of 'avoiding shared risks and avoiding bifurcation of our two systems' is carefully calibrated: market-oriented rather than security-focused, and leaves room for cooperative rather than confrontational engagement. The technical specificity of the agenda — explicitly covering both open and closed weight models — signals substantive rather than symbolic discussions. The timing of the NYC meetings days before the Trump-Xi summit on September 24 gives them direct diplomatic stakes, and Bessent's role as a trusted intermediary between Chinese officials and the US private sector makes him an effective point person for navigating that terrain.
Sentiment
Mostly neutral reporting with skepticism over confirmation and uncertainty
“Per @Reuters: US and China planning to discuss AI safety in mid-September, led by Treasury Secretary Scott Bessent... Proposed agenda involves cooperation on monitoring AI-directed cyberattacks, concern about a future Chinese 'Mythos-level' model... Craig Mundie is sounding out Chinese counterparts.”
“Reuters: mid-September US–China AI safety talks are coming. White House: no such meeting is currently planned. Ahead of Trump–Xi on Sep 24, even the process channel is contested.”
“NEWS: SOURCES SAY US CHINA PLAN MID SEPTEMBER AI TALKS... A WHITE HOUSE OFFICIAL SAID THERE IS CURRENTLY NO PLANNED AI RELATED MEETING IN MID SEPTEMBER... TRUMP AND XI ARE DUE TO MEET SEPTEMBER 24...”
Split
Split between those sharing the Reuters report as a positive diplomatic step (~60%) and those highlighting the White House denial and uncertainty (~40%).
Sources
- Exclusive-US, China gear up for mid-September AI safety talks - Yahoo FinanceFinance
- Sources: the US and China will discuss AI safety risks during talks planned for mid-September, with Treasury Secretary Scott Bessent leading the US sideReuters
- Sources: at the September 24 US-China talks, the US is expected to discuss preventing AI-directed cyberattacks, and China will likely revisit US export controlsAsia
- Scoop: U.S. open to discuss AI "shared risks" with China, Bessent says - AxiosAxios