NVDB Warns of Security Backdoors in Anthropic's Claude Code AI Programming Tool

China’s Network Security Threat and Vulnerability Information Sharing Platform (NVDB), under the Ministry of Industry and Information Technology, released a risk alert today pointing out security backdoor risks in the AI programming tool Claude Code.

According to NVDB’s monitoring results, Claude Code—developed by U.S.-based AI company Anthropic—can autonomously complete code writing and repair tasks based on text requests. However, the tool includes a built-in monitoring mechanism that transmits sensitive information like user geographic location and identity identifiers to remote servers without user consent. The affected versions are Claude Code 2.1.91 to 2.1.196.

NVDB advises relevant organizations and users to immediately conduct comprehensive checks: for development terminals with the affected versions installed, uninstall or upgrade to the latest secure version that has removed the backdoor code;同时, strengthen control over external connection permissions and traffic monitoring for development tools within core business networks to prevent unauthorized data leakage.

This security backdoor issue is linked to a previously exposed hidden detection mechanism. In late June 2026, developers reverse-engineering Claude Code 2.1.196 discovered that since the 2.1.91 version released on April 2, 2026, the tool has included a detection mechanism that checks system time zones and proxy server information to identify users related to China. This mechanism was never mentioned in subsequent update logs. Thariq Shihipar, a member of Anthropic’s Claude Code team, later responded on social media that the mechanism was an “experimental” measure launched in March 2026 to prevent unauthorized account resale and model distillation attacks. The team stated that the detection function was removed in the new version released on July 2, 2026.

IT Home noted that shortly before NVDB’s risk alert, Chinese tech enterprise Alibaba announced internally in early July 2026 that it would fully ban employees from using Claude Code in the office environment starting July 10, listing it as a high-risk software.

Sources

• Weibo (URL: https://m.weibo.cn/detail/5323268535094545)
• IT Home