Alibaba Bans Claude Code Over Hidden Chinese User Tracking and Security Risks

Alibaba Group Holding has issued an internal directive prohibiting its workforce from utilizing Anthropic's Claude Code software development assistant. The restriction is scheduled to take effect on July 10, 2026. According to an internal memo circulated by the Chinese technology giant, Claude Code has been classified as high-risk software possessing security vulnerabilities and potential back-door risks. The company has instructed its software developers to transition to its proprietary coding agent platform, Qoder, instead. Neither Alibaba nor Anthropic immediately responded to requests for comment regarding the directive.

The Steganographic Tracking Mechanism Uncovered

The workplace ban follows discoveries made by independent security researchers who analyzed the inner workings of Claude Code. On June 30, a Reddit user operating under the handle LegitMichel777 reverse-engineered the software and found hidden, obfuscated tracking code. This tracking mechanism had been silently integrated into the platform since version 2.1.91, which was published on April 2, without any acknowledgment in the official release documentation. The hidden code was designed to inspect the user's local system environment. Specifically, it checked if the system timezone was configured to Asia/Shanghai or Asia/Urumqi and cross-referenced active proxy URLs against a hardcoded index of Chinese web domains and artificial intelligence research laboratories.

Rather than transmitting this diagnostic data openly, the tool employed steganography to conceal the telemetry inside the system prompt sent to Anthropic's servers. For users flagged as operating from a Chinese timezone, the software altered the date format from dashes to slashes and substituted the standard apostrophe in the phrase "Today's date is" with one of three visually identical but technically distinct Unicode characters. This allowed Anthropic's servers to parse the geographic origin of the query without alerting the user or the artificial intelligence model itself.

Anthropic Defends the Experiment Amid Distillation Feud

In response to the public disclosure, Anthropic representatives addressed the underlying intent of the code. Thariq Shihipar, an engineer at Anthropic, stated on the social media platform X that the tracking mechanism was an experiment launched in March. According to Shihipar, the feature was designed to prevent account abuse from unauthorized resellers and protect against distillation. He noted that the tracking system would be rolled back.

The tracking controversy is closely linked to an ongoing intellectual property conflict between the two firms. Just last month, Anthropic accused Alibaba of orchestrating a massive distillation strike, which involves training a less advanced model on the outputs of a more capable one to illicitly extract its capabilities. Anthropic claimed this was the largest known attack of its kind on its systems, designed to help accelerate China's progress toward achieving capabilities equivalent to Anthropic's Mythos Preview model. While Claude Code has gained significant popularity among software developers in China, Anthropic officially restricts access to its tools for users and entities based in the country. Lizzi Lee, a fellow at the Asia Society Policy Institute's Centre for China Analysis, noted that if a United States-based AI coding tool can detect Chinese usage or proxy access, it is entirely expected that major Chinese technology firms would restrict its internal use.

This escalating cycle of covert tracking and retaliatory bans suggests that the boundary lines of the global AI race are no longer just being drawn by national trade policies, but are now actively coded directly into the software itself.