China Alleges 'Backdoor Risks' in Anthropic's Claude Code, Alibaba Bans Use

China's National Vulnerability Database (NVDB) has issued a severe warning against specific versions of Anthropic's AI coding tool, Claude Code, alleging the presence of "security backdoor risks" capable of transmitting sensitive user information without consent. The government-backed cybersecurity platform, affiliated with the Ministry of Industry and Information Technology, claims versions 2.1.91 through 2.1.196, released between April 2 and June 29, 2026, contain a hidden mechanism that sends user location and identity data to remote servers.

This alarming revelation stems from a security researcher's discovery of obfuscated code designed to detect users operating from China, particularly by checking system time zones and proxy URLs against a list of Chinese domains and AI labs. Anthropic engineer Thariq Shihipar confirmed the code was part of an "experiment" initiated in March 2026 to combat unauthorized reselling and protect against model "distillation," asserting that stronger mitigations are now in place and the controversial code was removed in a July 1st update.

The incident significantly escalates the ongoing US-China technological rivalry, particularly in the critical AI sector. Anthropic has previously accused Chinese entities, including Alibaba, of illicitly "distilling" its AI models. In response to the backdoor allegations, Chinese tech giant Alibaba has already instructed its employees to cease using Claude Code by July 10, opting instead for its in-house Qoder platform. The NVDB has urged all affected users and organizations to immediately uninstall or upgrade to the latest secure version and enhance network traffic monitoring to prevent unauthorized data leakage. This episode underscores the profound challenges of trust, intellectual property protection, and national security in the increasingly fractured global AI landscape.

Watch (Shorts)