Alibaba Bans Anthropic Claude Code Over Covert User Tracking Features

Alibaba Restricts Claude Code Citing Backdoor Risks

Alibaba Group Holding has instructed its employees to stop using Claude Code, the artificial intelligence coding assistant developed by United States-based firm Anthropic. According to an internal notice issued by the Chinese technology giant, the software will be banned from workplace use starting on July 10, 2026. Alibaba designated Claude Code as a high-risk application with security vulnerabilities, pointing to concerns that the tool had been used to secretly monitor users located in China. The restriction represents a significant shift for developers in China, where Claude Code has maintained strong popularity despite Anthropic officially restricting access to users and entities within the country. Employees have been directed to transition to Qoder, Alibaba's proprietary coding agent platform.

Steganography and Timezone Inspections

The security concerns erupted after independent researchers discovered hidden tracking mechanisms within the software. On June 30, a Reddit user operating under the handle LegitMichel777 reverse-engineered Claude Code and identified obfuscated tracking routines. This hidden code had been active since the release of version 2.1.91 on April 2, without any disclosure in the official release notes. The tracking program inspected user environments by checking if system timezones were set to Asia/Shanghai or Asia/Urumqi, while also scanning proxy URLs against a hardcoded directory of Chinese domains and artificial intelligence laboratories. Instead of generating standard log files, the software utilized steganography to secretly transmit data back to Anthropic. When a Chinese timezone was detected, the system altered the date format sent in prompts from dashes to slashes. Additionally, the apostrophe in the phrase "Today's date is" was swapped with one of three visually identical but technically distinct Unicode characters. This allowed Anthropic to parse the signals programmatically while keeping them invisible to human users.

Distillation Accusations and Defensive Experiments

The discovery of the tracking code follows a period of intense friction between the two technology firms. In June 2026, Anthropic sent a letter accusing Alibaba of executing an industrial-scale distillation attack on its models, calling it the largest known operation of its kind. Anthropic claimed that Alibaba was illicitly extracting capabilities from Claude to accelerate its own progress toward matching Anthropic's advanced Mythos Preview system. Addressing the tracking controversy on social media, an Anthropic employee stated on X that the covert tracking mechanism was an experiment launched in March. According to the employee, the experiment was designed to identify unauthorized resellers abusing accounts and to defend Anthropic's intellectual property against ongoing model distillation attempts.

Whether this defensive tracking represents a legitimate security measure or a dangerous escalation in corporate espionage, it highlights how deeply geopolitical distrust is now hardcoded into the very developer tools driving the artificial intelligence race.