September 18, 2026 — Technology, privacy and the Great Firewall
A dispute over ZCode, the AI coding desktop application developed by Zhipu AI, has raised a broader question for users and companies working inside mainland China’s controlled digital environment: when software quietly moves sensitive data to cloud infrastructure, how much visibility and control do users actually have?
The controversy began after developer Ferstar published a reverse-engineering analysis alleging that, while a user is logged in, ZCode can package an entire workspace — including source code, portions of Git history, LFS data, reflogs and configuration material — encrypt the resulting archive and send it to Alibaba Cloud OSS. Ferstar also reported that the encryption public key was supplied by the service while the corresponding private key remained on the server side.
Those findings are allegations based on one technical investigation and subsequent community attempts to reproduce the behavior. They should not be treated as proof that every ZCode installation uploaded the same categories of files, or that Zhipu retained or examined every uploaded archive.
Zhipu acknowledges uploads linked to Repo Wiki
Zhipu responded on September 18 by apologizing to affected users and saying the issue originated in its “code repository indexing” system. According to the company, the Repo Wiki feature could trigger repository-data uploads while generating cloud-based Wiki pages. Zhipu said the feature had initially been enabled by default, that the problem had been fixed, and that uploaded data used for the Wiki function was destroyed after the page was generated rather than retained.
The company also said it would open-source the ZCode codebase, invite third-party reviewers to examine the system and publish further audit information.
What remains unresolved
The public response does not yet resolve every technical question raised by the reverse-engineering report. In particular, independent reporting noted that Zhipu had not publicly reconciled the developer’s claim that full Git-history material could be included in local snapshots with the company’s description of filtered repository data used for Repo Wiki.
That distinction matters. Sending task-relevant code to a cloud AI service is expected behavior for many coding assistants. Quietly packaging a much broader repository history, if confirmed, would raise separate questions about informed consent, source-code confidentiality, trade secrets, data minimization and whether users can meaningfully disable or delete the transfer.
Why this belongs in the Great Firewall archive
Free China Movement’s Great Firewall & Internet Persecution in Mainland China project documents more than website blocking. It tracks the wider digital-control environment in mainland China: censorship, real-name systems, platform monitoring, data collection, device and account scrutiny, surveillance, and the human consequences of attempting to communicate outside controlled channels.
The ZCode dispute does not by itself prove that the software was operating as a government surveillance tool. However, FreeChina.live considers the observed architecture relevant to a potential state-surveillance pathway: if sensitive repository material is uploaded to cloud infrastructure and remains decryptable server-side, the technical conditions exist for third-party access. That concern is heightened by the PRC legal environment, under which Article 35 of the Data Security Law requires relevant organizations and individuals to cooperate when public-security or national-security organs lawfully obtain data. This is a risk assessment based on observed data flow and legal access authority, not evidence that authorities actually requested, received or used ZCode uploads.
The episode also demonstrates why technical transparency matters. Users need to know what data leaves a device, what triggers transmission, where the data is stored, who can decrypt it, how long it is retained, whether upload can be disabled, and whether those claims can be independently audited.
What users and organizations should verify
Organizations using AI coding tools should independently review outbound network connections, data-retention terms, enterprise controls and repository exclusions before allowing proprietary or sensitive code to be processed. Sensitive credentials found in repository history should be rotated if exposure is suspected. Claims about software behavior should be tested against current versions because vendors can change implementation rapidly after disclosure.
Sources
- Ferstar — “Inside ZCode: Silently Uploading Your Entire Git History to the Cloud”
- IT之家 — ZCode response and remediation announcement
- DoNews — reporting on the technical allegations and Zhipu response
- FreeChina.live — Great Firewall & Internet Persecution in Mainland China
Related FreeChina.live research
- Great Firewall of China: Technical History, Censorship Mechanisms, Cases, Data and Circumvention
- China Information Control & Circumvention History: Policies, Cases, Technologies and Countermeasures
Research note: The ZCode case is tracked in these projects as a cloud-data-flow, consent and platform-transparency issue. It should not be treated as proof of a Great Firewall mechanism or of government surveillance without additional evidence.
Evidence note: This article distinguishes the developer’s technical allegations from Zhipu’s own response. FreeChina.live will update the record if an open-source release, independent audit or additional reproducible evidence clarifies the disputed upload scope.