China has set stringent conditions for artificial intelligence (AI) discussions with the United States, expressing sharp criticism of American AI company Anthropic and emphasizing the need for clear regulations before meaningful dialogue can take place. The comments were made in a Sunday post by Yuyuantantian, a social media account linked to Chinese state broadcaster CCTV known for signaling official policy stances.
The post stated that the U.S. must demonstrate that its AI companies adhere to safety, disclosure, and auditing standards comparable to those expected in China. It called for a joint agreement among all parties involved to delineate “genuine security threats” from mere technological competition. “This line must be drawn jointly by all participating parties,” the message said, underscoring China’s insistence on mutual understandings before advancing talks.
The AI discussions are anticipated to precede Chinese President Xi Jinping’s scheduled state visit to the U.S. on September 24, where he is set to meet with President Donald Trump. However, no date has yet been announced for the AI talks, and Chinese officials reportedly remain uncertain about the agenda and the appropriate representation for the meetings.
Tensions have escalated between the two largest economies over AI leadership. The Yuyuantantian post singled out Anthropic’s Claude AI model for specific reproach, alleging that the model has breached user data boundaries, monitored users covertly, and transmitted website domains without permission. The post accused U.S. frontier AI models of developing along a “distorted direction” and characterized upcoming negotiations as continuing disputes over safety and regulatory standards rather than purely technical exchanges.
According to the account, the U.S. government has not adequately enforced security controls around AI technology and has “repeatedly retreated” from this responsibility. The piece titled “Anthropic Has Contracted the American Disease” claimed that the American approach amounts to an attempt to impose safety rules defined primarily by U.S. companies, which China finds unacceptable.
The post further framed the U.S. strategy as part of a broader effort to contain China’s technological ascent, criticizing American proposals as attempts to compel China into accepting an order shaped in part by U.S. corporate interests.
Concerns within Chinese official circles extend to Anthropic’s Mythos AI model, which some officials fear could be used offensively against China. Sources familiar with internal deliberations have noted Chinese unease over the company’s refusal to provide China with access to certain AI systems for legitimate purposes.
Anthropic blocked Chinese-controlled companies from accessing its services last year due to U.S. national security concerns, though Chinese firms appear to have continued using its products despite the restrictions. Anthropic, alongside OpenAI and Alphabet Inc.’s Google, has voiced worries about the use of their AI models in China. The company has also accused Alibaba Group Holding Ltd. of illicitly accessing Claude via thousands of fraudulent accounts, an allegation that Alibaba has not publicly addressed.
Requests for comment sent to Anthropic and the U.S. State Department outside of U.S. business hours were not immediately answered.