Trump-Xi Sept 24 summit to include AI safety talks, per Nikkei

AI safety is expected on the agenda when Presidents Trump and Xi meet in Washington on Sept 24, with the US side planning to raise the risk of AI-directed cyberattacks, Nikkei Asia reported.

A separate working-level dialogue is scheduled for mid-September, led on the US side by Treasury Secretary Scott Bessent, who has called for guardrails and closer communication between the 'world's two leading AI powers.' The Chinese delegation may be headed by Vice Premier He Lifeng or Ding Xuexiang. On the agenda: monitoring AI-directed cyberattacks and a proposal for US and Chinese labs to police themselves and share threat information.

Both sides bring a concrete rap sheet into the room. The White House has accused Moonshot AI of distilling Anthropic's Fable model to produce its K3 release, and in July nearly 700 malicious agents built on OpenAI models compromised Hugging Face. DeepSeek's V4 Pro is estimated to trail leading American models by around eight months, while Z.ai's GLM-5.2 performs comparably on coding and agent tasks.

Chip export controls are the piece the two sides disagree on airing at all. Commerce Secretary Howard Lutnick, speaking at G20 meetings in North Carolina on Sept 2, said he did not expect the Trump-Xi meeting to cover technology controls: 'I don't think so.' On H200 sales to China he added, 'The Chinese haven't really taken it up...So I don't think it's necessary.'

The dialogue would be the first dedicated US-China AI channel under Trump's second term, and it is not fully locked in. A White House representative said 'there is no AI-focused meeting planned for mid-September at this moment,' leaving agenda and attendees in motion. The broader chip fight has kept our tracker busy, with 344 China-AI stories logged in the last 90 days, most recently a NYT report that Inspur's US subsidiary Aivres shipped $3B in Nvidia Blackwell chips to Southeast Asia despite the ban.