China’s National Vulnerability Database issued a formal security advisory directing Chinese software developers to uninstall Claude Code versions spanning three months of releases, citing a “built-in monitoring mechanism” that the state-run database characterizes as unauthorized collection of user location and identity information. Alibaba announced a company-wide ban on Claude Code use for all employees effective the following day — the largest corporate enforcement action tied to the CNVDB advisory and the most significant AI developer tool prohibition in China since scrutiny of U.S. AI products intensified.
What the CNVDB Advisory Claims and Which Versions It Targets
The CNVDB advisory covers Claude Code versions 2.1.91 through 2.1.196, the range released between April 2 and June 29. The advisory does not direct users to update to a patched version — it instructs developers to uninstall the affected software entirely.
Which Developers Fall Within the Three-Month Version Range
The affected version range encompasses three months of routine Claude Code releases, meaning developers who incorporated Claude Code into their standard workflow throughout Q2 used versions within the advisory’s scope. Any developer in China who installed Claude Code during this period and did not subsequently remove it is running software the CNVDB has formally advised against. The breadth of the advisory — covering the tool rather than a specific vulnerability class — is consistent with how Chinese government advisories handle foreign software products where the concern is data handling practice rather than an exploitable bug.
Alibaba’s Employee Ban as the First Major Corporate Enforcement of a CNVDB AI Advisory
Alibaba’s decision to prohibit Claude Code use for all employees, effective the day after the CNVDB advisory, represents the first documented case of a Chinese enterprise translating a CNVDB AI tool advisory into a company-wide prohibition. The speed of the enforcement — announced the same day the advisory published — suggests Alibaba had reviewed the advisory and acted before most organizations had an opportunity to evaluate it. The ban applies to all employees regardless of role or project, making it a blanket restriction rather than a targeted policy for sensitive projects.
Anthropic’s Public Dispute of the “Backdoor” Characterization
Anthropic publicly responded to the CNVDB advisory by disputing the characterization of the mechanism as a backdoor. Anthropic described the feature CNVDB labeled surveillance infrastructure as an experimental anti-abuse measure launched to protect Claude Code from unauthorized resellers and to counter model distillation — a practice where competing firms systematically collect input-output pairs from legitimate usage to reverse-engineer the capabilities of advanced AI models.
Anthropic’s framing describes a mechanism intended to protect the company’s intellectual property and enforcement compliance, not to surveil users. The CNVDB’s framing — “built-in monitoring mechanism” transmitting user location and identity to remote servers — describes the same mechanism from the perspective of what it technically does rather than what Anthropic intended it to do. Both descriptions can be accurate simultaneously: an anti-distillation measure may involve collecting usage metadata that a foreign government regulatory body characterizes as unauthorized surveillance under Chinese data law.
How the CNVDB Claude Code Advisory Fits China’s Pattern of U.S. AI Tool Scrutiny
The Claude Code advisory follows an established pattern in which Chinese government bodies apply formal scrutiny to U.S.-origin AI products that collect usage data or operate with opaque data-handling mechanisms. The CNVDB serves as the state apparatus through which China evaluates and advises on foreign software security, and the Claude Code advisory represents the body exercising that mandate against an AI coding tool with significant adoption among Chinese software developers.
What the Advisory Reveals About AI Tool Bifurcation Along Geopolitical Lines
The practical effect of the advisory — combined with Alibaba’s enforcement action — is that Claude Code becomes effectively unusable as a professional development tool for Chinese software engineers at major enterprises subject to compliance requirements. Developers at Alibaba and any organizations that follow the advisory face a choice between CNVDB compliance and continued access to the tool. The advisory does not create a path to compliant use through a patched or modified version; the prescribed action is removal. For developers who built Claude Code into their workflow throughout the affected version window, the advisory requires a tool replacement decision rather than a routine update.
