China’s National Vulnerability Database (NVDB) has identified a potential security vulnerability in Anthropic’s coding assistant, Claude Code, urging developers to uninstall or update specific versions. The flagged versions—ranging from 2.1.91 to 2.1.196—are said to contain a built-in mechanism that transmits sensitive user information, including region and identity data, to external servers without user consent.
This alert follows Alibaba’s recent decision to prohibit its employees from using Claude Code, citing concerns over data security and potential identification of China-linked information. The NVDB, operating under China’s Ministry of Industry and Information Technology, did not disclose technical details on how the security flaw was discovered but emphasized the risk posed by these versions in corporate environments.
Developers within China are advised to remove affected versions from development systems immediately and to transition to updated software iterations that address these vulnerabilities. Additionally, organizations are encouraged to monitor coding tools connected beyond internal networks to prevent unauthorized data transmissions.
The warning emerges amidst heightened tensions around AI technology access in China. Anthropic previously stated that it does not commercially license Claude within China due to national security concerns, yet acknowledged that some Chinese entities have circumvented restrictions using proxy accounts. This situation echoes broader debates over embedded backdoors in technology platforms, with U.S. hardware and software likewise facing scrutiny over surveillance vulnerabilities.
No official response from Anthropic has surfaced regarding these allegations. Meanwhile, the incident underscores increasing scrutiny on AI tools’ data handling practices and the risks of unregulated information flows in software used in sensitive development settings.