Why Alibaba Is Completely Banning Anthropic’s Claude Models

Alibaba evaluated Anthropic’s Claude series and, citing a security risk that could embed a backdoor, added the models to its high‑risk software list. The company then mandated that every employee uninstall Claude Sonnet, Claude Opus, Claude Fable, Claude Code and related proxy tools by July 10.

Earlier in the year, Alibaba encouraged staff to adopt AI technologies, offering free internal model quotas and reimbursing external model usage. Engineers freely chose Claude, GPT, Gemini and other models, spending hundreds of dollars per week, and frequently used Claude Code, OpenAI Codex and Alibaba’s own Qoder as agent tools.

In February, Anthropic accused three Chinese AI firms—DeepSeek, Moonlight, and MiniMax—of stealing Claude technology via distillation. By June, Anthropic alleged that Alibaba employed roughly 25,000 fraudulent accounts to interact with Claude over 28.8 million times, extracting data through personnel linked to Alibaba and its Tongyi Lab. Anthropic reported these claims to the U.S. Senate Banking Committee, elevating the issue to a national‑security concern.

Reacting to being listed as a “Chinese military enterprise” by U.S. authorities, Alibaba filed a lawsuit in the San Jose federal court seeking removal from the 1260H blacklist.

From late June to early July, Anthropic rolled out a blocking campaign that abruptly cancelled many Chinese users’ personal and team subscriptions without refunds. Claude Code now employs technical checks to detect whether an agent or API address references Chinese companies such as Alibaba, Ant Group, or ByteDance, and subtly flags Chinese users in system prompts.

As of today, Alibaba cites these security concerns as the reason for a complete ban on Claude across the organization.