China raises AI safety concerns ahead of expected talks with US

China has accused US AI companies of operating under inadequate safeguards and said Beijing will require stronger assurances before engaging in substantive AI talks with Washington.

Yuyuantantian, a social-media account affiliated with CCTV, said the US should first prove that American AI companies are governed by the same safety, disclosure and auditing standards that Washington expects China to follow. It argued that security threats should be clearly separated from technological competition and that boundaries should be established jointly.

The commentary specifically targeted Anthropic, alleging that Claude had crossed user-data boundaries, conducted covert monitoring and transmitted website domains without authorization. It also accused Washington of failing to adequately enforce AI security boundaries while attempting to make standards shaped by US companies into global norms.

The rhetoric comes ahead of expected US-China AI talks, which are being discussed in the run-up to Xi Jinping’s planned Sept. 24 summit with Donald Trump. The timing and agenda of the AI dialogue have not yet been finalized.

The two countries are increasingly competing over advanced AI models and infrastructure. Chinese officials have reportedly worried that Anthropic’s Mythos could be used offensively against China, while Anthropic has blocked services for Chinese-controlled companies over US national-security concerns and accused Alibaba of illicitly accessing Claude through fraudulent accounts.