A new challenge from China has pushed Anthropic’s Claude Code into the center of a wider argument over AI safety, competition, and trust. On August 31, 2026, Yuyuantantian, a social media account affiliated with Chinese state broadcaster CCTV, published a lengthy critical commentary titled “Anthropic Has Contracted the American Disease.”
The post does more than criticize one AI coding tool. It lays out conditions for US-China AI talks, argues that Anthropic overstepped user data boundaries, and calls for closer scrutiny of how American model companies identify and handle sensitive information.
Two Conditions for AI Talks
The commentary says US-China AI discussions must separate genuine security threats from ordinary technological competition. That distinction forms the first part of its argument: capabilities that could cause serious harm should be tested and limited together, while normal model research and commercial use should not face covert interference.
The second condition focuses on whether the United States applies its proposed safety rules to its own companies. The post asks whether Washington should investigate American firms, publish the purposes, scope, and rules of their identification mechanisms, and submit those mechanisms to third-party audits.
Only after the US proves that safety rules bind American model companies, the commentary argues, can the two sides enter substantive discussion. That demand puts transparency at the heart of any future talks, connecting technical safeguards with the conduct of the companies building advanced AI systems.
The message is direct: AI competition should not become a cover for hidden interference, but tools that create serious security risks should face joint testing and limits. The commentary uses Claude Code as its main example of where it believes that line may have been crossed.
Why Claude Code Became the Focus
Much of the post targets Claude Code, Anthropic’s AI coding tool. China’s Ministry of Industry and Information Technology named Claude Code more than a month ago, alleging that it carried serious security backdoor risks by transmitting sensitive information without consent.
Yuyuantantian’s commentary cites discussions among developers on GitHub to support its criticism. In January 2026, an overseas developer reviewed proxy logs and found Claude Code sending requests to a website every few seconds. Other developers reported that the tool transmitted site domains to Anthropic’s interface before user authorization.
Those claims place data handling at the center of the dispute. A coding tool that connects with outside systems can raise questions about what information leaves a user’s environment, when those transfers happen, and whether the user has authorized them. The commentary presents those questions as security concerns rather than routine model development issues.
The post also criticizes a parameter called “dangerously-skip-permissions.” That parameter lets Claude Code execute operations without confirmation, creating another point of concern in the commentary’s account of the tool’s behavior.
Auto Mode Adds Another Flashpoint
The commentary also mentions an “auto mode” introduced in March 2026. That mode was enabled by default from August 14, 2026, making its presence part of the post’s broader examination of how Claude Code handles permissions and user control.
Together, the allegations about requests, site domains, permission settings, and auto mode form the core of Yuyuantantian’s criticism. The post links these details to its larger demand that American AI companies face clear rules, open explanations, and third-party checks.
The argument now places Anthropic inside a much larger policy test. Can the US demonstrate that its AI safety standards apply to its own companies? Can both sides distinguish a real security threat from ordinary commercial competition? And can audits provide enough evidence for substantive talks to begin?
Yuyuantantian’s August 31 commentary offers its answer: discussions require proof before trust. With Claude Code at the center, the next stage of this dispute will turn on how the United States responds to demands for investigations, published identification rules, and independent audits.
