China has issued a security warning against Claude Code, the AI coding assistant developed by Anthropic, claiming that several versions of the software contain a hidden "backdoor" capable of collecting sensitive user information. According to China's National Vulnerability Database, the affected versions could transmit data such as a user's location and identity to Anthropic's servers without explicit consent, prompting officials to urge users to update or remove the software.
The controversy follows reports that the detection mechanism was designed to identify users in China who were accessing Claude Code through workarounds, despite Anthropic officially restricting its services in the country. An Anthropic engineer said the feature was part of an anti-abuse experiment intended to detect unauthorized access and prevent model distillation, rather than to spy on users. The company has since removed the mechanism after it became public.
The dispute has further intensified tensions between U.S. and Chinese AI companies. In response to the security concerns, Chinese technology giant Alibaba reportedly instructed employees to stop using Claude Code and instead switch to its in-house AI coding platform. The incident comes amid an increasingly competitive AI landscape, with both countries tightening controls over access to advanced AI technologies.
Security experts say the incident highlights the growing importance of transparency in AI software, especially for enterprise and government users. While China has characterized the feature as a serious security vulnerability, Anthropic maintains it was a security measure aimed at preventing abuse of its services. The episode underscores how geopolitical tensions are increasingly influencing the development, regulation, and adoption of AI tools worldwide.
