On September 18, 2026, a developer named ferstar published a reverse-engineering report revealing that ZCode, an AI coding desktop app by Beijing-based Z.ai, silently uploads users’ entire git workspace history to Alibaba Cloud’s object storage service, Aliyun OSS. The encrypted archive includes the full .git history, LFS asset cache, reflogs, and global app configurations, raising significant privacy concerns.
The researcher captured a 313MB encrypted archive created from a 345MB commercial workspace containing 42,411 files, with 564 failed upload attempts logged during the investigation. ZCode operates as a closed-source harness for the open-weight GLM models developed by Z.ai, which are widely used in the local AI community. Despite the open nature of the GLM weights, the ZCode app itself is proprietary, and the uploads occur silently whenever the app is logged in.
This revelation highlights the distinction between open-weight AI models and the proprietary runtimes built on top of them. The incident has caused confusion among users, some of whom mistakenly believed ZCode was open source due to its use of GLM models. The silent uploading of entire git histories to a cloud service without explicit user consent raises questions about data privacy and security in AI development tools, especially those integrating large language models.
The report by ferstar was published on September 18, 2026, on tokenstead.ai, sparking discussions in both English and Chinese developer communities about the implications of ZCode’s data handling practices.