Anthropic Responds to China Cybersecurity Alert Over Claude Code

Washington, July 9: Anthropic pushed back on Thursday against claims by a Chinese government-run cybersecurity database that the artificial intelligence company’s Claude Code developer tool had a security vulnerability allowing operators to access users’ data surreptitiously, saying users in China never had access to the tool.

The Chinese database, run by the Ministry of Industry and Information Technology, said Wednesday in a statement on its WeChat account tied to the National Vulnerability Database that Claude Code had a “backdoor” that could transmit users’ location data and other identifying information to a third-party server. It said impacted versions, which ranged from 2.1.91 to 2.1.196 and included models released from early April to late June, should be either uninstalled or updated, calling the bug a “significant risk” to Chinese users.

Anthropic spokesperson said Thursday that Anthropic does not allow use of Claude Code by people in or citizens of China or by companies headquartered in China. As such, “Claude Code has never been available to Chinese users,” she said. The versions affected by China’s advisory have since been updated. On Thursday, version 2.1.204 was the latest iteration available for download on Anthropic’s website.

Claude Code gained traction in China even without official access because Anthropic hasn’t launched the tool there or in Hong Kong. In March, a Xiaomi engineer told attendees at a government-run event that developers were already using Claude Code within China, according to a video transcript reviewed by CNBC.

Last month, Anthropic said it purposefully included a tracker in Claude Code earlier this year as part of an experiment to prevent “distillation,” a practice where companies take prompts and responses from a leading AI model to train smaller models themselves. Companies including Chinese giants DeepSeek, MiniMax and Alibaba Group Holding have quietly distilled Anthropic’s models, Anthropic alleged in a blog post.

An Alibaba spokesperson declined to comment when CNBC and other media first reported the distillation claims in May.

This month, Alibaba warned employees in a memo obtained by CNBC and the South China Morning Post that it would ban workers from using Claude Code beginning July 10 due to security concerns. The ban stems from the alleged backdoor, according to the Alibaba memo.

Claude Code is just one of several recent sticking points between Anthropic and Chinese tech companies. Last month, Alibaba and fellow Chinese AI companies DeepSeek and MiniMax debuted AI chatbots less than two weeks after Anthropic threatened legal action over allegedly distilling its Claude model.