China issues national red alert against Anthropic's Claude Code over "Backdoor" Claims

China’s industry ministry has issued a national red alert against Anthropic’s AI coding tool, Claude Code. The government mandate requires domestic organizations and individual developers to immediately audit their digital environments and uninstall specific software builds, alleging the presence of a malicious "backdoor" that covertly siphons sensitive user data to remote American servers without authorization.

The emergency security warning was published by the National Vulnerability Database (NVDB) - a state-run platform operating directly under the Ministry of Industry and Information Technology (MIIT). According to the formal advisory, the security risk targets Claude Code versions 2.1.91 through 2.1.196, encompassing a wide suite of updates deployed throughout the spring season.

Regulators claim these versions feature a built-in monitoring mechanism capable of quietly collecting and transmitting persistent identity-related identifiers and precise geographic location data.

Spying on the Spies to Protect Core Models

The red alert lands just weeks after a fierce corporate dispute broke out between Anthropic and Chinese e-commerce giant Alibaba. Anthropic formally accused Alibaba’s Qwen AI team of orchestrating the largest "distillation attack" in history, deploying 25,000 fraudulent accounts to run 29 million automated exchanges with Claude to reverse-engineer its proprietary reasoning logic and train its own domestic models.

The alleged "backdoor" flagged by Chinese authorities was actually the tracking code Anthropic deployed to catch those thieves. Anthropic engineer Thariq Shihipar confirmed the script was an anti-abuse experiment meant to unmask unauthorized regional proxy resellers and guard against illegal model copying.

Enforcing a Domestic Software Quarantine

To ensure the American tool is completely purged from critical infrastructure, the NVDB directive has urged Chinese enterprises to implement aggressive network-level containment strategies. In addition to forcing standard uninstalls, organizations are being directed to tighten outward-bound firewall rules, block external web connections originating from third-party developer suites, and significantly ramp up deep-packet traffic monitoring on core business servers to intercept unauthorized outbound data bursts.

The regulatory dragnet has already triggered swift corporate compliance. Alibaba officially placed Claude Code on its high-risk corporate blacklist, strictly barring all employees from utilizing the American coding tool for any software development tasks. Employees have instead been ordered to transition their workflows entirely to Alibaba's native, state-approved internal alternative, Qoder.