China Demands US AI Lab Audits for AI Talks, First Terms Ahead of September 24 Xi Visit

A CCTV-affiliated account demanded the US implement binding regulations and third-party audits for frontier AI labs before substantive bilateral talks can begin. The requirements include granting the US government power to investigate labs and publicizing classified benchmarks for assessing cyber risk, according to a Yuyuantantian article. The account specifically targeted Anthropic, accusing the company of overstepping user data boundaries and conducting covert monitoring.

The demands arrive three and a half weeks before President Xi Jinping's September 24 visit to the US. Beijing also requested a joint definition to distinguish security threats from technological competition to prohibit US restrictions on Chinese users' access to frontier models. US Treasury Secretary Scott Bessent previously said he would pursue "very robust" AI discussions focusing on preventing powerful models from reaching non-state actors.