China’s Ministry of Industry cybersecurity platform warned on Wednesday that it had identified a so-called “backdoor” in Anthropic’s code authoring tool, Claude Code. According to the Chinese authorities, the software contains a built-in monitoring feature that can transmit sensitive information without users’ consent.
The notice named the affected releases as Claude Code versions 2.1.91 through 2.1.196. The operators of the National Vulnerability Database posted on their WeChat channel that companies and users should immediately inspect their systems, remove vulnerable versions, or update to the latest secure release, which they say no longer contains the disputed code. They also urged organisations to tighten controls on external network access for developer tools and to increase monitoring of data flows on critical business networks to prevent unauthorised leakage of confidential information.
Corporate reactions and Anthropic’s response
According to a Reuters report, Alibaba has already banned employee use of Claude Code at work after features in the software raised concerns that it could be used to identify users linked to China.
Anthropic rejected the backdoor allegation, saying the cited code is an experimental abuse-prevention mechanism and not a covert data exfiltration channel. The company also noted that Claude services are not officially available in China.
Why this matters
The episode highlights security risks associated with developer tools and AI-driven services, and it underscores the importance of network access controls and data-monitoring practices for organisations. Because the Chinese authority’s claim and Anthropic’s description differ, further technical scrutiny and potential software updates are likely. Meanwhile, corporate decisions to restrict or remove the tool may affect Claude Code’s enterprise usage.
(Source: Reuters.)
