Alibaba will ban Claude Code from its employees after detecting hidden functions to identify Chinese users

Alibaba will prevent that your employees use Claude Code de Anthropic starting next Friday, July 10, due to security concerns after identifying internal code in the model that, as ‘spyware’, automatically identified its use by Chinese users.

The artificial intelligence (AI) company has already expressed on several occasions its intention to prevent other foreign AI companies from accessing their most capable modelswith special emphasis on Chinese companies, which he accuses of carrying out distillation attacks using Claude to improve their models.

In this context, it has been announced that Anthropic published a Claude Code version that integrated hidden code with which the model could identify when it was being used by a user in Chinawithout the person knowing.

This has been shared by a user named LegiMichel777 through Reddit, who has verified that, from the Claude Code version 2.1.9.1published in April, the model can identify if the user has a ‘proxy’ activated and, through “invisible modifications to the system message”, determine whether it is being used in China. All of this, invisible to the user himself.

This hidden code practice has been confirmed by Anthropic worker, Thariq Shihipar, in a publication on the social network This is an “experiment” that the company launched on Claude Code during the month of March with the aim of “Prevent account abuse by unauthorized resellers and protect against distillation”.

Specifically, he clarified that the team “has implemented stronger mitigations since then” and that its intention is “deactivate” this experiment for a while now. However, he concluded by ensuring that this practice “should be completely reverted” in the new version published by Claude Code.

However, after knowing these facts, since Alibaba have determined that Claude Code is now “high risk” software for national security, and has decided prohibit its use by its employees starting next Friday, July 10as detailed by sources familiar with the Chinese AI company and reported by media such as Morningstar and TechCrunch.

It is worth remembering that Anthropic recently denounced a massive operation of Alibaba to replicate Claude’s technology by using around 25,000 fake accounts, who accessed the model to copy its technologies using distillation techniques, in order to train their own Qwen AI models at a lower cost.