ZCode, the GLM coding agent, silently uploads your Git history (tokenstead.ai)

On September 18, 2026, developer ferstar revealed that ZCode, an AI coding agent from Beijing's Z.ai, surreptitiously uploads users' entire Git histories to Alibaba Cloud's Aliyun OSS whenever logged in. This alarming discovery has raised significant privacy concerns within the AI/ML community, particularly among users of locally-run models like GLM-5.3-Flash. The ZCode app creates an encrypted archive of a user’s workspace, including sensitive files and potentially exposed internal data such as API keys and unreleased product plans, all without user consent. The incident underscores the importance of understanding the implications of the runtime environment of AI tools, where closed-source applications may not offer the same transparency and security as their open-source counterparts. The upload mechanism relies on envelope encryption, where the data is encrypted with a symmetric key, wrapped by a public RSA key stored on Z.ai's servers, making it impossible for users to decrypt their own data. Although Z.ai touted transparency during ZCode's launch, the practice of covertly packaging and transferring extensive user data is a significant breach of trust. The viral nature of this discovery, with over 276,000 views on ferstar's post within hours, has prompted widespread discussions about the need for due diligence regarding the data handling practices of AI tools, emphasizing the necessity for users to scrutinize runtime behaviors and encryption policies.