AnswersPublished27 sources
What ZCode’s Repo Wiki Uploaded—and What Z.AI Changed Afterward
Researchers reported that ZCode’s default Codebase Indexing/Repo Wiki workflow could package an encrypted snapshot of a logged in developer’s workspace—including Git history—and send it toward Alibaba Cloud OSS. The key unresolved issue is evidentiary: Z.AI and its commissioned assessments say the relevant OSS bucke...
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer
Key takeaways
- Researchers reported that ZCode’s default Codebase Indexing/Repo Wiki workflow could package an encrypted snapshot of a logged in developer’s workspace—including Git history—and send it toward Alibaba Cloud OSS.
- The key unresolved issue is evidentiary: Z.AI and its commissioned assessments say the relevant OSS bucket was cleared, but an empty bucket after remediation cannot by itself establish the full prior processing histor...
- Taiyuan Chengming Technology subsequently demanded a processing inventory, deletion evidence, and answers on sharing, cross border transfer, and model training use—showing why verifiable data handling, not merely a pa...
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How did Z.AI’s ZCode coding assistant secretly package and attempt to upload developers’ local workspace files through its default Codebase. Article summary: ZCode’s default Codebase Indexing/Repo Wiki feature appears to have created and attempted to send an encrypted snapshot of an entire logged-in developer workspace—not merely files selected for an AI prompt—to Alibaba Clo. Topic tags: general, general web, user generated, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts wi
ZCode’s data-handling controversy was not simply about sending the files a developer selected for an AI prompt. A reverse-engineering report by researcher ferstar alleged that, while a user was logged in, the desktop client could package an encrypted workspace snapshot and initiate transfer to Aliyun (Alibaba Cloud) Object Storage Service. Z.AI acknowledged a related issue in its Codebase Indexing feature, saying Repo Wiki generation could trigger repository-data uploads during an early period when the feature was enabled by default.2
3
5
What researchers said the client packaged
According to ferstar’s technical report, the snapshot could encompass the whole workspace rather than a narrow set of prompt attachments. Reported contents included:
- the complete
.gitdirectory and commit history; - Git LFS asset cache and reflogs;
- workspace files; and
- global application configuration.
3
5
One report described a roughly 313 MB encrypted archive. Reporting based on the analysis also said the relevant RSA private key was held on Z.AI’s side, which would prevent a user from independently decrypting a captured archive or using the ciphertext alone to verify deletion.4
These are reported findings from reverse engineering and traffic observation, not a public, independent inventory of every affected machine. The distinction matters: evidence of a transfer attempt to OSS does not establish that every attempt completed, nor does it by itself show how long an object remained available after transfer.3
16
How Repo Wiki was connected to the upload behavior
Z.AI said the issue originated in Codebase Indexing, a feature used for capabilities including Repo Wiki and version-related workflows. Its account was that generating a Repo Wiki page could trigger the repository-data upload, and that the feature had been on by default in the early release period.2
5
11
That explanation addresses the product pathway, but it does not make the behavior equivalent to a conventional, explicit file-upload action. The central trust issue was scope and consent: users and researchers described a full-workspace package being created without a clear, effective confirmation for that transfer.3
6
Z.AI’s remediation: removal, open source, and audits
After the reports emerged, Z.AI apologized and disabled affected features.2 The company later said that ZCode v3.14.0 removed Repo Wiki and disabled the workflow for generating and uploading local repository snapshots.
16
Z.AI also said it open-sourced ZCode and invited external scrutiny of the client. Reports described assessments by the China Academy of Information and Communications Technology (CAICT) and NSFOCUS.10
23
In ZCode’s published account of those assessments:
- CAICT found the
zcode-prodAlibaba Cloud OSS bucket in a zero-data state after remediation; and - NSFOCUS confirmed deletion of data objects in the bucket and related backup storage.
16
Separately, Zhipu said it planned a no-data-retention option for its MaaS platform, under which inputs and outputs would be used for the current request and deleted afterward. That proposal concerns the platform’s stated future processing policy; it should not be treated as independent proof of what happened in the earlier ZCode workflow.24
What remains unverified
The company’s remediation claims are meaningful, but they do not answer every historical question by themselves. A post-remediation assessment showing an empty named bucket supports the claim that the reviewed bucket was cleared at that time. It cannot, without broader logs and a complete processing record, independently prove that data was never copied elsewhere, accessed before deletion, or used for another purpose.
Z.AI has said the reported code and related data were not retained or used for model training.41 However, the public material summarized here does not provide an independently auditable, end-to-end record that resolves that negative claim for every reported snapshot. The appropriate conclusion is therefore limited: the company denies training use, while public verification remains incomplete.
Why Chengming Technology requested a processing inventory
Taiyuan Chengming Technology sent Zhipu a legal demand after alleging unauthorized uploads of company data and trade secrets. Its reported requests went beyond deletion: it sought identification of storage locations, potential cross-border transfer, third-party sharing, possible model-training use, access or export logs, and proof that data had been deleted from servers, caches, backups, and disaster-recovery systems.17
21
26
That request identifies the practical standard users increasingly expect from AI coding products: not only a statement that data was deleted, but a traceable account of what was collected, where it traveled, who could process it, and how deletion can be verified.
Market reaction was immediate but not a lasting verdict
Reports said Z.AI shares fell during trading after the controversy; one report described an intraday decline of as much as 5%.35 Available closing quote pages also showed a 1.79% gain at a later quoted close, while another market report described a lower close for the day.
34
38 Those mixed snapshots support the narrower conclusion that the incident produced short-term market concern, not a reliable conclusion about long-term valuation impact.
The developer takeaway
The incident highlights a basic evaluation criterion for AI coding tools: distinguish local indexing, prompt-context transfer, and full-workspace snapshot upload. A product should make each behavior legible, separately controllable, and easy to verify through documentation, network inspection, or source review.
For teams handling proprietary code, secrets, or regulated data, the safer operational posture is to require explicit upload consent, minimize workspace scope, review outbound network destinations, and seek vendor commitments that specify retention, subprocessors, deletion procedures, and audit evidence. In this case, Z.AI’s removal of the snapshot-upload workflow addresses the reported mechanism, while the broader questions of historical data handling explain why the controversy continued after the patch.16
17
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "What ZCode’s Repo Wiki Uploaded—and What Z.AI Changed Afterward"?
Researchers reported that ZCode’s default Codebase Indexing/Repo Wiki workflow could package an encrypted snapshot of a logged in developer’s workspace—including Git history—and send it toward Alibaba Cloud OSS.
What are the key points to validate first?
Researchers reported that ZCode’s default Codebase Indexing/Repo Wiki workflow could package an encrypted snapshot of a logged in developer’s workspace—including Git history—and send it toward Alibaba Cloud OSS. The key unresolved issue is evidentiary: Z.AI and its commissioned assessments say the relevant OSS bucket was cleared, but an empty bucket after remediation cannot by itself establish the full prior processing histor...
What should I do next in practice?
Taiyuan Chengming Technology subsequently demanded a processing inventory, deletion evidence, and answers on sharing, cross border transfer, and model training use—showing why verifiable data handling, not merely a pa...