An account affiliated with Chinese state broadcaster CCTV has published a blistering critique of Anthropic while laying out conditions for US-China AI safety talks. The commentary, titled “Anthropic Has Contracted the American Disease,” appeared on Yuyuantantian, an account widely read as a signal of official policy thinking.
The post argues that frontier model controls, and who gets to define them, are the central dispute between the two countries. It sets two preconditions for substantive negotiation. First, a line must be drawn jointly between genuine security threats and ordinary technological competition, with capabilities that could cause serious harm tested and limited together.
Second, it says Washington must prove its proposed rules bind its own companies, asking whether US firms should first face investigations, published disclosure of their identification mechanisms, and third-party audits before China is pressed to restrict model releases.
Much of the attack targets Claude Code, Anthropic’s coding tool. China’s Ministry of Industry and Information Technology named it more than a month ago, alleging serious security backdoor risks, according to the post. The commentary lands as the two governments prepare a new round of AI discussions.
