Security News

Z.ai's Coding Tool Caught Secretly Uploading User Code, Company Apologises

The Register · 23 Sept 2026

Key Takeaway Before adopting AI coding tools, check their privacy policies and settings for any data upload or telemetry features, and avoid using them on sensitive or proprietary codebases until you can verify how your data is handled.

Chinese AI company Z.ai has apologised after researchers discovered its code-generation tool, ZCode, was packaging up entire user workspaces, including full project histories, and sending them to Alibaba Cloud without users' knowledge. The behaviour echoes a similar controversy involving Elon Musk's xAI earlier this year.

Researcher Ferstar, who first flagged the issue, found there was no setting to disable the uploads and no mention of the practice in ZCode's privacy policy. Making matters worse, the decryption key for the uploaded data was held only by Z.ai's own servers, meaning affected users had no way to access or delete their own files once uploaded. The company traced the problem to a feature called Repository Index, which triggered uploads after generating documentation pages in the cloud.

Z.ai says the uploaded data was never used to train its AI models. It has since removed the offending feature, commissioned two external security reviews confirming the data was deleted, and open sourced the ZCode project on GitHub for public scrutiny. However, Ferstar noted that Z.ai wiped commit history and the original source code that performed the uploads, limiting independent verification of what actually happened.

Summarised by CISO AI from The Register. We link back to every original so you can read it yourself.