Alibaba has told staff to stop using Claude Code, effective today, classifying it as high-risk software and directing employees to its own internal tool, Qoder, instead. The trigger was a Reddit post claiming the tool contained covert functionality to identify users’ location and network setup. Anthropic’s response, from engineer Thariq Shihipar, was that the flagged behaviour was “an experiment we launched in March that was meant to prevent account abuse from unauthorized resellers and protect against distillation” — an anti-abuse safeguard, not covert tracking — and that stronger mitigations have since replaced it.
Whether or not the original concern held up, the episode is a useful data point on its own. AI coding tools now sit deep enough inside company workflows — reading source code, running commands, touching credentials — that a single misread telemetry feature can trigger an outright corporate ban within days, reported globally, at one of the world’s largest tech employers. That’s a different risk profile to a normal developer tool, and it’s one that’s going to get more scrutiny, not less, as adoption grows.
So what
This is worth raising directly with any team standardising on an AI coding assistant, rather than treating it as a settled choice once it’s in the toolchain: what does the tool send back to its vendor, under what circumstances, and who’s checked. It’s a governance question as much as a technical one, and it belongs in the same conversation as data residency, GDPR compliance, and access control on anything client-facing. If you’re weighing AI-assisted development tooling for a regulated or security-conscious build, our AI-assisted development page covers how we handle this, or get in touch to talk it through.