Starting July 10, Chinese tech giant Alibaba is imposing a complete ban on the use of Anthropic products by its employees in work environments. As an alternative, employees are encouraged to switch to its own Qoder platform. The restrictions cover all Claude models, including Sonnet, Opus, and Fable, as well as the developer tool Claude Code. The company has officially classified this software as "high-risk."

Technical Incident as a Trigger

The reason for such a drastic decision lies in a recent security incident. Developers discovered that version Claude Code 2.1.91, released in April 2026, contained a hidden mechanism for checking the user's environment. The tool read the local time zone and checked proxy server addresses and API endpoints for keywords related to Chinese cloud providers and AI companies. If a match was found, it altered the date format and punctuation in the system prompt without sending explicit telemetry.

Tariq Shihinah, a member of the Claude Code team, explained that this was a "March experiment" to combat unauthorized account resale and protect against model distillation. He promised that the mechanism would be removed in the next release. However, for Alibaba, this explanation proved insufficient.

Geopolitical Context and Shadow Economy

Alibaba's ban is not an isolated event. It is a direct consequence of Anthropic's tightening policies towards China. As early as September 2025, the company updated its sales rules, excluding mainland China and Hong Kong from the list of supported regions due to risks of data transfer to intelligence agencies. Later, Anthropic accused Chinese startups DeepSeek, Moonshot, and MiniMax of a large-scale campaign to use Claude to improve their own models, recording over 16 million interactions from 24,000 fraudulent accounts.

In June, Anthropic repeated its accusations, this time against Alibaba and its AI division Qwen, calling it the largest known distillation attack. According to insiders, it was this letter to the U.S. Senate that became the trigger for the internal ban at Alibaba.

Despite the strict restrictions, demand for Claude in China remains high. As investigations show, a shadow market for accounts thrives on Taobao and Telegram, while intermediary stations in Singapore generate abnormally high traffic for model usage. Researchers note that Chinese models still lag behind their American counterparts by 6-9 months, especially in coding tasks, which drives this black market.

Expert opinion: This incident is a clear marker that the era of "free" AI has ended. We are witnessing the formation of digital iron curtains in the field of artificial intelligence. Technical incidents, like the one discovered in Claude Code, are merely becoming pretexts for political and corporate decisions. The AI tools market is fragmenting, and in the coming years, we will see the emergence of fully isolated ecosystems, which will inevitably slow global technological progress but create new niches for local players like Qoder.