Alibaba bans staff from Claude Code over backdoor fears

TL;DR:

  • Alibaba has banned employees from using Anthropic’s Claude Code at work after developers found features that can help identify China-linked users.
  • The ban deepens a dispute in which Anthropic has accused Alibaba of “distilling” its Claude models to accelerate rival systems.
  • Staff are being directed to Alibaba’s own coding platform, Qoder, as Chinese firms pivot to domestic and open-source models.

The move, reported by Reuters, hardens a corporate fault line running straight through the US-China AI rivalry. Claude Code, Anthropic’s coding assistant, had become popular among Chinese developers despite the company’s restrictions on access from China. Developers recently found the tool inspected user environments — including timezone and proxy information — and inserted subtle markers into prompts sent to Anthropic’s servers. An Anthropic employee described the feature on X as “an experiment we launched in March” to prevent account abuse by unauthorised resellers and guard against model distillation.

Trust becomes a compliance question

Anthropic said last month it had suffered a “distillation” strike by Alibaba — training a weaker model on the outputs of a stronger one — which it claimed was helping China approach its advanced Mythos Preview capabilities, according to a letter seen by Reuters and sent to two US senators. Alibaba has not publicly addressed the accusations.

For businesses, the episode is a reminder that AI tooling now carries geopolitical baggage. Enforcement against individual users is hard — a developer can route traffic through US servers — but companies are more alert to legal and compliance exposure, the Reuters source noted. The result is a decoupling: Chinese cloud and AI firms are shifting toward DeepSeek, Alibaba’s own Qwen, Moonshot and Zhipu, even as cheaper Chinese models close the gap with US leaders.

Looking forward

The standoff sharpens a question every enterprise buyer now faces: how much do you trust the provenance and telemetry of the AI tools your developers use? For UK firms operating across both markets, the practical lesson is due diligence on what coding assistants collect and transmit — and an awareness that the tools themselves are becoming instruments in a broader contest for AI leadership.

Share this article