Z.ai apologized for training on code without permission and open-sourced ZCode after security vulnerabilities were discovered. The incident highlighted concerns about AI companies' data practices similar to issues previously seen with other AI systems.
Z.ai announced plans to open-source its ZCode codebase following backlash over data storage practices. The company said it stored user data in its cloud without a disable option and committed to third-party audits and transparency to rebuild trust.
Reverse-engineering analysis revealed that ZCode, an AI coding app from Beijing-based Z.ai, silently packages and encrypts users' entire Git history and workspace data, uploading it to Aliyun OSS with encryption keys held only by Z.ai's servers. The discovery sparked privacy concerns across developer communities, with critics noting the closed-source harness contradicts assumptions about openness tied to Z.ai's open-weight GLM models.
ZCode, an AI coding desktop app from Beijing-based Z.ai, silently uploads users' entire Git history and workspace files to Aliyun OSS whenever logged in, with encryption keys held only by Z.ai's servers, preventing users from decrypting their own data. A reverse-engineering analysis by ferstar revealed the practice captures 86.6% git object stores containing deleted API keys, unpushed branches, and internal hostnames, with UI toggles failing to stop the uploads.
ZCode, Zhipu's AI coding desktop app, secretly uploads users' entire workspace including Git history and project files to Aliyun OSS when logged in. The encrypted archives are encrypted with server-held RSA keys that users cannot access locally, meaning the ciphertext on disk cannot be decrypted by the user or client.