Alibaba will bar employees from using Claude Code, Anthropic’s coding agent, starting July 10, according to multiple reports cited by TechCrunch on July 4. The Chinese company has reportedly classified the tool as high-risk software and will direct staff toward Qoder, its own coding assistant, instead. The move formalizes a divide that has been building for months, since Anthropic already blocks Chinese companies, and foreign entities those companies own, from using its models at all.
Anthropic has also been working to close loopholes that let Chinese users reach Claude despite that ban. According to a Reddit post cited by TechCrunch, part of that effort involved a version of Claude Code built to secretly identify Chinese users. Anthropic’s Thariq Shihipar addressed the practice on X.
Shihipar described it as “an experiment we launched in March that was meant to prevent account abuse from unauthorized resellers and protect against distillation,” a reference to the practice of training one model on the outputs of another. He said the team had since landed stronger mitigations and had been meaning to retire the check regardless of the Reddit thread.
That explanation surfaced only after a Reddit user found the code and asked about it. Anthropic did not disclose the detection feature when it shipped in March. The sequence matters: the company explained itself only under public pressure, not ahead of it.
The high-risk label signals what Alibaba’s security team weighed most heavily: exposure of proprietary code and internal architecture to a rival’s infrastructure. Coding agents route large volumes of a company’s codebase through the vendor’s servers by design. That is how they generate useful completions. A vendor already caught building undisclosed detection logic for a specific set of users becomes a harder vendor to trust with that exposure.
Distillation risk cuts in Alibaba’s favor too. Anthropic has spent months restricting Chinese access specifically to prevent distillation, training cheaper models on a frontier model’s outputs. Alibaba develops its own frontier models under the Qwen brand, giving it a direct incentive to keep employee prompts and completions off a rival lab’s servers, independent of whatever Anthropic itself intended.
The dispute illustrates a broader split forming in enterprise AI tooling. Anthropic restricts access by the user’s nationality and corporate ownership. Alibaba restricts access by the vendor’s nationality and its own risk classification. Multinational teams now operate inside two separate approval processes for the same category of software, and the criteria behind each are not published in detail.
For builders selling coding agents into enterprise accounts with Chinese operations, the practical test just moved from feature parity to data governance. Confirm today which tool touches proprietary code before an employer or a regulator forces the choice for you.
Reported by TechCrunch on July 4, 2026.
