China Accuses Claude Code of Hidden Backdoor, Alibaba Bans Tool for Employees

China's Ministry of Industry and Information Technology (MIIT) announced on July 8 that its cyber threat detection platform had found a security vulnerability in Claude Code, Anthropic's coding tool, describing it as a backdoor. According to the ministry, the tool was able to send a user's location and identity data to a remote server without their consent.

The Chinese ministry's accusation came just days after Alibaba became the first company to ban its employees from using Claude Code, citing reports of a hidden location-detection mechanism targeting China. The company instructed its teams to switch to Qoder, an in-house coding tool developed by Alibaba itself.

What the ministry and researchers found

According to accounts cited by tech media, the mechanism in Claude Code reportedly checked whether the system was running in the Shanghai or Urumqi time zone, then compared the proxy address against a hardcoded list of domains linked to Chinese companies and AI labs, including Alibaba, Baidu, Ant Group and ByteDance. Rather than sending an explicit signal, the result of that check was said to be encoded invisibly to the user, through subtle changes in date formatting and punctuation in the response sent back to Anthropic's servers.

This setup would have allowed Anthropic to identify whether the tool was being used by someone linked to the Chinese market or a rival AI lab, without disclosing this to the user. The Chinese ministry called the mechanism a serious security threat and recommended that users uninstall or update the affected versions of the tool.

Anthropic's response

Anthropic does not deny the mechanism exists, but rejects calling it a backdoor. Thariq Shihipar, who works on Claude Code, wrote on X that the code was part of an experiment intended to prevent abuse by unauthorized resellers and to protect against model distillation.

The team has since rolled out stronger protections, and we'd actually been planning to remove this for a while - Thariq Shihipar, Claude Code engineer at Anthropic

The company stresses that, under its policy, China and other countries designated as adversarial are not permitted to use Anthropic's models at all, which it says puts the whole matter in a different light than a conventional leak of data from Western users.

Background of the Alibaba dispute

Tensions between Anthropic and Alibaba have been building for months. In a letter to the leadership of the U.S. Senate Banking Committee on June 10, Anthropic accused entities linked to the Qwen lab of running industrial-scale attempts to distill Claude's software engineering and reasoning capabilities, using close to 25,000 fake accounts to do so. Alibaba's ban on employees using Claude Code, effective July 10, is both a response to those accusations and a counterattack leveraging reports of the alleged backdoor.

What it means for Polish companies

For Polish companies using Claude Code, the affair is mainly a matter of reputation, and a reminder of the risk posed by hidden telemetry mechanisms in AI coding tools. The mechanism in question was designed to detect specific time zones and domains linked to Chinese entities, so its direct impact on users outside that circle appears limited, but the dispute shows how geopolitics is shaping the architecture of popular developer tools.

Companies operating in regulated industries should pay attention to what telemetry data the AI tools used by their development teams send out, and should demand clear information from vendors about the anti-abuse mechanisms built into their products.

The whole affair is unfolding against the backdrop of a broader US-China trade and technology conflict in artificial intelligence, where accusations of intellectual property theft and hidden control mechanisms are flying in both directions.

Sources: China warns of security backdoor in Anthropic AI coding tool (cbsnews.com), Alibaba bans Anthropic's Claude Code after an alleged hidden China-detection backdoor is uncovered (tomshardware.com), China's Alibaba bans Anthropic AI for employees after distillation attack accusation (cnbc.com)