- Employees will be barred from using Anthropic’s coding tool from July 10, with Qoder designated as the replacement
- The move follows reports that Claude Code contained code capable of identifying users in China through hidden system checks
Alibaba Group will prohibit employees from using Anthropic’s Claude Code in its internal working environment beginning July 10 after classifying the software as high risk following reports that it contained hidden code capable of identifying users in China.
Employees have been instructed to switch to Alibaba’s in-house coding assistant Qoder, according to an internal notice.
The ban extends to Anthropic’s broader model lineup, including the Sonnet, Opus and Fable series.
Targeted at Chinese entities
The decision follows disclosures that Claude Code version 2.1.91 contained code that, when a proxy was enabled, checked whether a user’s system was set to China’s time zone and compared the proxy URL against a list of 147 domains associated with Chinese technology companies and AI laboratories.
According to reports circulating among developers, the software then embedded the detection result into system prompts through subtle changes such as date formatting without notifying users.
The mechanism reportedly remained in the software for nearly three months before being discovered.
Against distillation attacks
Anthropic engineers later acknowledged the feature, describing it as an experimental measure introduced in March to combat account abuse and model distillation attacks.
The company said the code was completely removed in a new version released on July 2.
The episode follows escalating tensions between the two companies.
On June 24, Anthropic submitted a letter to the U.S. Senate Committee on Banking, Housing and Urban Affairs accusing Alibaba of using about 25,000 fake accounts to conduct more than 28.8 million conversations with Claude, alleging an industrial-scale model distillation campaign.
Alibaba’s internal ban marks a notable shift in its AI policy.
Reducing reliance on Anthropic’s models
The company had previously subsidized employees’ use of external AI models, reimbursing access to services including Claude and GPT alongside free quotas for its own models. That policy has now been reversed.
Other Chinese technology companies have also reduced their reliance on Anthropic’s models following changes to the company’s terms of service.
After Anthropic announced in September 2025 that Chinese-controlled companies were prohibited from using its services, ByteDance removed Claude models from its Trae coding assistant.
Likewise, Tencent replaced Claude in the international version of CodeBuddy with alternatives from OpenAI and Google.
However, neither ByteDance nor Tencent has issued a company-wide internal ban comparable to Alibaba’s.
Industry analysts said Alibaba’s decision could prompt broader reviews of overseas AI tools by Chinese tech firms.
