Alibaba is preparing to ban its employees from using Anthropic's Claude Code, the AI coding assistant that has become a default tool across large parts of the software industry, a source told Reuters. The move formalises a break that has been building for months and hands Alibaba's engineers back to domestic alternatives, chiefly the company's own Qwen-based coding stack.
The ban applies to internal use of Claude Code across Alibaba's engineering teams. Alibaba has not publicly detailed the policy, and the exact enforcement mechanism — network-level blocks, expense-policy denials, or both — has not been disclosed. The Reuters report cites a single source familiar with the matter.
Anthropic set the stage for this in September, when it restricted access to its models for majority-Chinese-owned entities and their overseas subsidiaries, citing national-security concerns tied to potential misuse and to US export-control obligations. That policy effectively closed the front door for Alibaba to run Claude through Anthropic's API in the first place. What is changing now is that Alibaba is also closing the back door, telling staff not to route around the restriction via personal accounts or third-party resellers.
Key facts
- 01Alibaba plans to ban employees from using Anthropic's Claude coding tool internally, according to a source cited by Reuters.
- 02The move follows Anthropic's earlier decision to cut off Chinese-owned entities from its models on national-security grounds.
- 03Alibaba operates its own Qwen family of models and a competing coding assistant, giving it a domestic fallback.
- 04The ban lands as Washington lifts export curbs on Claude Fable 5 and Mythos 5 for allied buyers, sharpening the US-China split.
Claude Code has been one of Anthropic's fastest-growing product surfaces, competing directly with GitHub Copilot, Cursor, and a wave of agentic coding tools. For a hyperscaler-scale engineering organisation, losing access to it is not trivial: Claude's Sonnet and Opus models sit at or near the top of most public coding benchmarks, and the internal productivity delta between Claude Code and second-tier assistants is measurable.
Alibaba, for its part, is not starting from zero. Its Qwen model family — Qwen 3 and the Qwen-Coder variants in particular — is one of the strongest open-weight Chinese model lineages, and Alibaba Cloud ships coding assistants built on top of it. The ban therefore doubles as a mandate: use the in-house stack, improve it under real internal load, and stop feeding usage data and dependency to a US rival that has already declared you off-limits.
The timing is pointed. Washington last week lifted export curbs on Anthropic's Claude Fable 5 and Mythos 5 for allied buyers, a signal that the US is willing to let its frontier labs sell abroad — just not to Chinese firms. That asymmetry is the backdrop to Alibaba's move. If Chinese hyperscalers cannot buy Claude, and cannot let their engineers use it unofficially, they have to build the alternative themselves and sell it into markets Anthropic cannot serve.
This fits a broader pattern across the Chinese tech sector. ByteDance, Tencent, and Baidu have all invested heavily in in-house foundation models and coding tools, in part because access to US-made frontier models has become unreliable as a matter of policy. DeepSeek's rise this year — its open-weights V3 and R1 releases in particular — has given Chinese engineering teams a credible domestic option that did not exist eighteen months ago. Alibaba banning Claude Code is less a shock than a ratification of a bifurcation that already exists in practice.
There are open questions. Alibaba has not confirmed the policy on the record, and the source's characterisation could be narrower than a blanket ban — for instance, limited to code touching sensitive infrastructure, or to specific business units. Anthropic has not commented on Alibaba's internal decisions. And enforcement inside a company of Alibaba's size, where hundreds of thousands of employees use dozens of tools, is a real operational problem regardless of intent.
The move confirms that the AI tooling market is splitting cleanly along the US-China line, and that the split now reaches down to the developer-productivity layer, not just to model weights and chips. For Anthropic, losing Alibaba as even an unofficial user base is a rounding error on revenue but a meaningful signal about total addressable market: China is closed. For Alibaba, the calculation is the mirror image — every engineer forced onto Qwen-Coder is a data point that improves the domestic stack and reduces the cost of the next decoupling shock.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
