TL;DR — Key Takeaways
- China is setting strict preconditions for upcoming AI safety talks with the U.S., arguing that American AI companies should face stronger safety, disclosure and auditing requirements.
- Beijing has singled out Anthropic, accusing the company of data misconduct, covert monitoring and using technology restrictions to reinforce U.S. dominance.
- Chinese officials are also raising concerns about the possible offensive use of Anthropic’s Mythos model against Chinese targets.
China has adopted a defiant stance ahead of upcoming artificial intelligence (AI) safety discussions with the United States, issuing strict preconditions for dialogue while directly accusing Anthropic of data misconduct and technological containment.
In a Sunday post published on Yuyuantantian, an influential state media account affiliated with state broadcaster CCTV, Beijing signaled that Washington must subject its domestic AI developers to rigorous safety, disclosure, and auditing standards before any “substantive” negotiations can take place.
The commentary emphasized that international guardrails must not be dictated unilaterally by Washington, warning that “a clear distinction must be drawn between genuine security threats and mere technological competition.”
Neither Anthropic nor the U.S. State Department responded to requests for comment.
The sharpening rhetoric comes ahead of anticipated bilateral AI talks intended to precede Chinese President Xi Jinping’s scheduled state visit to the U.S. on Sept. 24 for a summit with President Donald Trump. While officials have agreed in principle to deliberate on AI safety, Beijing has continued to push for clarity regarding the official agenda and delegation requirements.
Central to China’s grievances is Anthropic. State media explicitly accused the company of “distorted” development practices, citing alleged user data violations, unauthorized domain transmissions, and covert monitoring. Beijing frame these operational practices as part of a broader strategy to impose American tech dominance as global standard baseline rules.
Tensions surrounding high-stakes AI technology have escalated as both nations weigh the offensive and defensive capabilities of advanced frontier models. Chinese officials have expressed concern over Anthropic’s Mythos model being potentially deployed as an offensive tool against Chinese targets. Meanwhile, Beijing continues to query why access to such models remains restricted for routine commercial applications in China.
Anthropic blocked access to its services for Chinese-controlled entities last year on national security grounds — a move shared alongside similar regulatory precautions by competitors OpenAI and Google parent Alphabet Inc.
Despite these restrictions, cross-border friction has persisted. Anthropic recently accused Chinese tech giant Alibaba Group Holding Ltd. of illicitly bypassing regional blocks to access its Claude model via thousands of fraudulent accounts, an allegation Alibaba has not publicly addressed.
As diplomatic arrangements for the September summit continue, state commentary warned that Washington’s regulatory posture reflects a persistent effort to constrain China’s technological rise, concluding that U.S.-led controls represent “an attempt to make China accept an order partly defined by American companies.”