China alleges that Claude Code contains backdoors, calls mechanism 'a serious threat' — Gov't claims Claude sends sensitive information to remote servers without consent
TL;DR
China’s NVDB warned against Claude Code versions released from April to June 2026, describing them as containing backdoor vulnerabilities. Users were told to uninstall or update. Affected builds allegedly could send location and identity data to remote servers through a built-in monitoring mechanism without user consent.
Nauti's Take
The word backdoor is loaded, but Anthropic made the trust problem worse for itself. A coding agent should not quietly inspect usage environment, domains, or location signals and explain only after public pressure that it was anti-abuse work.
Beijing is not a neutral security referee, yet the core point lands: agents with repo and terminal access need less hidden cleverness and more verifiable transparency.
Briefingshow
Coding agents often run inside projects, terminals, and repositories. If a vendor adds hidden telemetry there, a plausible anti-abuse rationale is not enough; companies need auditable disclosure, version control, and clear data flows. China is also using the case inside the wider fight over access to U.
S. AI tools and domestic alternatives.
Sources