Updated today·Day 7·first covered 17 Sep 2026·19 sources

Xi Jinping's Washington visit and AI diplomacy

The gist

US-China AI Incident Alert Agreed; Cooperation Faces Structural Obstacles

A functioning government-to-government channel during an AI crisis would require the two leading AI powers to define shared thresholds and procedures, work that has not yet happened. The agreement's scope excludes chips and broader technology competition, leaving the main structural tensions untouched.

The full picture

The US and China agreed to establish a bilateral AI incident notification channel, proposed during September 20 talks in New York between Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng. The mechanism would allow either country to alert the other when AI systems behave unpredictably or pose threats reaching a national-security threshold. Chip and semiconductor export controls were explicitly excluded from the arrangement. China's official readout acknowledged dialogues on AI-related issues but did not confirm acceptance of Washington's specific notification mechanism. No published bilateral technical specification yet defines notification thresholds, authenticated contacts, information-sharing procedures, or escalation rules. Xi Jinping arrived in Washington for a summit with Trump, but analysts cautioned that little AI-related diplomacy should be expected. CSIS concluded the summit revealed how little progress has been made on the most consequential dimensions of US-China competition, including AI, cyber operations, and export controls. Experts also warn that continuing export controls and a US Justice Department accusation of IP theft by six Chinese AI companies two weeks before the talks could undermine any real cooperation.

How it developed

23 September 2026

Ars Technica reports experts warn export controls and Justice Department IP-theft accusations against six Chinese AI firms may undermine cooperation

The Trump-Xi summit produced minimal movement on AI governance, cyber operations, and export controls, per CSIS, following Trump's September 22 UN address in which he pledged to encourage rather than restrict superintelligence and designated the Department of Justice as the oversight mechanism. AI labs seeking antitrust exemptions to coordinate on safety now face opposition from officials including Hawley, Vance, and Kratsios, who frame the request as a competitive reward rather than a coordination problem.

First citedZvi's AI Roundups Rohan Paul Twitter csis.org piie.com

22 September 2026

Analysis finds government officials blocking AI labs from seeking antitrust exemptions for safety coordination.

Sam Altman, Jensen Huang, Tim Cook, Mark Zuckerberg, and Dario Amodei were confirmed as expected state dinner attendees for Xi Jinping's Washington visit, and Treasury Secretary Bessent said the US is open to discussing AI 'shared risks' with China. Huang publicly rejected Amodei's proposal for a broad ban on American technology sales to China. RAND published a paper describing the current US posture as analogous to accelerating with no investment in safety, and recommended preserving strategic options instead.

First citedThe Neuron Transformer News Zvi's AI RoundupsSemafor Technology+7

21 September 2026

US and China agree to establish a dedicated AI incident communication channel; no technical specification yet published

19 September 2026

Perplexity CEO Aravind Srinivas argues export controls may inadvertently convert China into a stronger AI competitor

18 September 2026

Rep. Ro Khanna announces shadow congressional process, inviting DeepSeek and Alibaba to join a binding pacing agreement

17 September 2026

China's intelligence chief publicly names Claude and GPT-5.5-Cyber as weaponized models, shifting official threat framing

16 September 2026

Atlantic piece argues US rhetoric about an existential AI race is undermined by lax export controls and missing policy actions

15 September 2026