Alibaba tells employees to stop using Anthropic’s Claude Code

Alibaba Group Holding has issued a directive prohibiting its employees from utilizing Anthropic’s Claude Code, citing significant security concerns. The decision stems from revelations about the AI company's previous use of concealed code aimed at identifying users within China. Following an internal assessment, Alibaba has classified Claude Code as high-risk software, which will take effect across all its offices starting July 10. In a notice, Alibaba stated, "Due to recent findings indicating potential back-door risks associated with Claude Code, we have included it on our high-risk software list after a thorough evaluation." This action follows allegations that Anthropic had embedded tracking mechanisms in its AI coding tool to ascertain whether users were located in China or affiliated with Chinese AI laboratories. Security researchers highlighted these practices earlier in the week, sharing their findings on platforms like Reddit and GitHub. Reports indicated that the hidden code was capable of gathering data such as the user's time-zone settings and proxy server use, subsequently relaying this information back to Anthropic via undisclosed messages, as noted by the cybersecurity outlet International Cyber Digest. In response to the backlash, Anthropic engineer Thariq Shihipar took to X to clarify the situation, describing the tracking feature as an "experiment" initiated in March. He explained that its intention was to thwart misuse by unauthorized resellers and safeguard Anthropic's AI models from the distillation process—where the outputs of a larger AI model are leveraged to train a smaller, less expensive one. Previously, Anthropic had accused various Chinese AI firms, including Alibaba, of employing distillation to replicate the capabilities of advanced AI systems from the US. Shihipar confirmed that the tracking feature would be eliminated in an upcoming re-release of Claude Fable 5, indicating that stronger protective measures were being developed. This incident unfolds against a backdrop of increasing restrictions on the deployment of advanced US AI technologies abroad. Last month, the US government imposed export controls on Fable 5, the public version of Anthropic’s more sophisticated Mythos model, following the discovery of security vulnerabilities by Amazon researchers. Although Anthropic temporarily disabled access to the model for all users, access was restored after the restrictions were lifted. Amidst these developments, Alibaba's actions mirror a broader trend among Chinese tech companies to lessen their dependence on American software and AI systems. Employees have been advised to utilize Qoder, Alibaba's proprietary AI coding tool, as an alternative to Claude Code. Although Anthropic's products are not officially available in China, many Chinese developers continue to leverage them via third-party services, drawn by their advanced coding and reasoning capabilities.