China's MIIT advisory alleges data-exfiltration backdoor in Anthropic's Claude Code

China’s Ministry of Industry and Information Technology escalated its standoff with Anthropic on July 8, 2026, issuing a formal advisory that alleges the company’s Claude Code coding tool contained a data-exfiltration backdoor.

The advisory, published through the ministry’s national vulnerability platform, NVDB, said the affected versions, 2.1.91 through 2.1.196, carried a built-in mechanism that could transmit sensitive user data, including geographic location and identity-related identifiers, to remote servers without consent. It advised organizations to uninstall or upgrade affected versions immediately.

The move formalizes a controversy that began in late June, when developers reverse-engineered Claude Code and found a hidden detection routine, present since version 2.1.91 in April, that checked users’ timezone and proxy data to flag connections linked to China.

Anthropic frames the feature differently. Thariq Shihipar, an engineer at the company, said on social media it was “an experiment we launched in March that was meant to prevent account abuse from unauthorized resellers and protect against distillation,” and that it was removed in the July 2 release.

The advisory raises the stakes in a widening access dispute. Alibaba told employees the prior week it would ban internal use of the tool starting July 10 over the same concerns.

The allegation has not been independently verified, and the two sides describe the same code in opposite terms, Beijing as covert data exfiltration and Anthropic as anti-abuse enforcement. Neither has published the underlying advisory or a technical rebuttal.