A developer who reverse-engineered ZCode, the AI coding tool from Z.ai (the Chinese research lab behind the GLM models), reports that the application — while running on a user’s own machine — created snapshots of entire developer workspaces and uploaded them, encrypted, to Z.ai's cloud when the user was logged in. According to the report, the uploaded contents included full Git history, Git LFS files, local branches, and operation logs.
The reverse-engineering author gave a concrete example: one project upload totaled 345 MB, and the .git directory alone accounted for 86.6% of that data. The researcher states that no user setting disabled the uploads.
Z.ai's response and change
Z.ai characterized the behaviour as "indexing" and said the feature was enabled by default. The company later removed the functionality, according to the report. While the uploads were reportedly encrypted, the researcher emphasizes that encryption does not address the core issue that a locally running application nevertheless transferred developer files to the vendor's cloud without an explicit opt-in beyond being logged in.
Why this matters
The case highlights an operational and privacy concern with tools that run locally but still transmit data to a central service. Because the computational work occurred on the user's device, Z.ai avoided the cloud compute costs associated with ingesting and processing customers' repositories, yet the vendor still obtained copies of the files. That combination raises questions about business models which rely on local execution for cost savings while collecting user data for their own use.
Implications and recommendations
The incident is a reminder that local execution does not automatically equal complete privacy. Organizations and individual developers should audit the behaviour of locally installed tools, monitor network traffic, and verify default settings. Users of such tools should check what data is collected and transmitted and whether there are options to disable automatic indexing or uploads.
