Alibaba has ordered employees to stop using Anthropic’s AI coding tool Claude Code from July 10, classifying the product as high-risk software and directing staff to move to the company’s in-house development tool, Qoder. The restriction applies to Alibaba employees globally. Reuters and Dow Jones, cited in the source material, said the decision was tied to concerns over potential backdoor risk.
Secret user-identification mechanism drew scrutiny
The ban comes after growing tension around Anthropic’s controls on access from China. Anthropic had already barred Chinese companies and overseas entities owned by Chinese capital from using its models. Then a Reddit post pushed the issue into wider view, alleging that Claude Code contained a version capable of secretly identifying users in China, sparking accusations that the software behaved like spyware.
Anthropic’s Thariq Shihipar later said on X that the mechanism was part of an experiment launched in March. According to his explanation, the goal was to prevent abuse by unauthorized reseller accounts and protect the model from distillation attacks, a practice in which outputs from another AI model are used to train a separate model. That response did not end the dispute.
Conflict escalated after distillation allegations
The source outlines a longer timeline behind the break. In April, Anthropic publicly accused Alibaba of carrying out what it called the “largest cloning attack in history,” alleging at least 28.8 million API calls were used against Claude to distill the model’s capabilities. Alibaba later ordered employees to remove Claude products across the lineup, marking a sharper split between the two companies.
Once the China-user identification mechanism became public, the dispute moved beyond access restrictions and into software security concerns. Alibaba’s latest step formalizes that position by putting Claude Code on its high-risk software list and replacing it internally with Qoder.
AI decoupling is reaching developer tools
The episode also points to a broader shift in the US-China AI divide. The source links this move to earlier export restrictions on high-end NVIDIA chips and to access blocks imposed by companies such as OpenAI and Anthropic on Chinese IP addresses. The divide is no longer limited to chips. It is now extending into software products used in day-to-day development work.
Alibaba’s push for Qoder reflects a wider effort by Chinese technology firms to build domestic alternatives as access to foreign AI systems tightens. In this case, the coding assistant itself has become part of the supply-chain risk calculation.