Friday, 16 October 2026
1 agent hack today▼ 7 vs yesterday (8)

Z.ai apologises after ZCode uploaded user workspaces to the cloud

The Chinese AI firm removed the feature, commissioned two external audits and open-sourced its coding tool after developers found it shipping entire projects to Alibaba Cloud.

By The Agentic Times ·

Z.ai apologises after ZCode uploaded user workspaces to the cloud
· Image: theregister.com

Z.ai has apologised after developers found that ZCode, its code-generation harness, was packaging up entire user workspaces and uploading them to Alibaba Cloud without consent, The Register reported on 22 September 2026.

According to The Register, ZCode packaged and git-encrypted complete user workspaces, including full project histories, and shipped them to Alibaba Cloud storage. The private key needed to decrypt that data was held only by a server under Z.ai's control. That meant users could neither access the files ZCode had uploaded on their behalf nor delete them.

The researcher who first raised the issue, who goes by Ferstar, said there was no setting that allowed users to turn the behaviour off, and that the practice was not disclosed in ZCode's privacy policy. Ferstar said the root of the problem was the tool's Repository Index functionality, which triggered file uploads after the Repo Wiki feature generated pages in the cloud.

ZCode published a statement on Monday apologising for the "security issues" and confirming that the uploaded data had never been used to train its models. "We sincerely thank the community developers who previously identified issues in ZCode. Going forward, we will establish an ongoing product security vulnerability reporting and response process," the company posted. It added that it would pay rewards to developers who report further problems, scaled to severity.

Z.ai said it had asked the China Academy of Information and Communications Technology and the Beijing security company NSFOCUS to examine the product after the changes were made. The company said both external assessments concluded that all previously uploaded data has now been deleted, and that the Repo Wiki feature has been removed. It said the full security assessment report would be released soon.

The company also open-sourced the entire ZCode project on GitHub, saying this placed the code "under community scrutiny" and made the tool "more open and transparent". Ferstar confirmed that the published code showed no sign that Repo Wiki was still implemented, but criticised the company for wiping commit records and removing the source code ZCode had used to upload files before the fix.

The Register drew a comparison with scrutiny of Elon Musk's xAI in July over similar issues.

Z.ai, known internationally until recently as Zhipu, spun out of Tsinghua University's Knowledge Engineering Group research lab in 2019. It is among the most heavily backed large language model companies in China and, according to The Register, the first AI company of the post-generative-AI era to float on the Hong Kong Stock Exchange. Last month the company claimed its latest model, GLM-5.3, matched the most advanced equivalents from Anthropic and OpenAI at finding security vulnerabilities. Anthropic and OpenAI have reportedly raised concerns about models from Z.ai and Moonshot, while the US government considers restricting access to them.

Sources

  1. theregister.comhttps://theregister.com/security/2026/09/22/zai-says-sorry-for-slurping-up-your-code-open-sources-zcode/5298300