Chinese tech firm reportedly bans use of Anthropic's programming tool amid allegations of model distillation
According to informed sources cited by Reuters, a major Chinese technology giant has prohibited its employees from using Claude Code, an AI coding assistant developed by the U.S.-based artificial intelligence company Anthropic, during work. The tool had previously drawn attention due to certain features capable of identifying users linked to China.
The ban is said to be part of an escalating dispute between the Chinese tech giant and Anthropic. Previously, Anthropic accused the Chinese enterprise of unauthorized extraction of its Claude large model capabilities, highlighting the intensifying competition between Chinese and American companies over leadership in artificial intelligence technology.
Claude Code is Anthropic’s AI coding assistant designed for software developers. Despite Anthropic’s restrictions on Chinese users and institutions accessing its services, the tool remains popular among many Chinese programmers.
According to the sources, Chinese enterprises have instructed their staff to switch to Qoder, a domestically developed AI coding platform.
Anthropic Accuses Chinese Firm of "Model Distillation"
Anthropic stated last month that Chinese companies had attempted to perform "model distillation" on its models—using outputs generated by more powerful models to train weaker ones.
In a letter submitted to two U.S. federal senators, Anthropic claimed this practice helps accelerate Chinese companies’ ability to catch up with the advanced capabilities of its Mythos Preview model.
Just days before the ban was implemented, developers discovered that Claude Code contained mechanisms capable of detecting user environments—including timezone and proxy server information—and embedding hidden markers into prompts sent to Anthropic’s servers.
An Anthropic employee posted on social platform X earlier this week stating that the feature was part of an experiment launched by the company in March, aimed at preventing unauthorized account reselling and protecting models from distillation.
The sources also noted that while restrictions targeting Chinese users are difficult to enforce for individual users—since they can easily route access through U.S. servers to make traffic appear as originating from America—enterprise users are far more sensitive to legal and compliance risks.
As U.S. AI model developers continue strengthening defenses against unauthorized access, reselling, and model distillation, Chinese cloud computing and AI firms are increasingly turning toward domestic and open-source models.
Meanwhile, Chinese AI models are gradually entering the U.S. market, raising concerns among some American industry professionals.
Chinese media had previously reported on the implementation of these bans by Chinese enterprises.
Source: rfi
Original article: toutiao.com/article/1869759577953353/
Disclaimer: This article represents the views of the author(s) alone