The data upload controversy surrounding ZCode, an AI coding tool developed by Zhipu AI, is now moving toward a resolution.
On Monday, Zhipu AI announced that ZCode had completed its technical remediation. All cloud data involved in the incident has been deleted, with the deletion verified by two third-party organizations, the China Academy of Information and Communications Technology and NSFOCUS. Zhipu also announced a compensation plan for all users, including free Token credits.
The data security incident, which was first brought to light by the developer community on September 18, has continued for more than 10 days, prompting changes to the product, the release of its source code, and third-party security checks.
The controversy started with a 313MB file
The incident began when developer ferstar was cleaning up files on a local computer. On September 18, the developer discovered an encrypted file of around 313MB in ZCode’s local data directory. Reverse engineering indicated that the file was generated by packaging the user’s entire workspace while logged in.
It contained project source code, Git history, LFS file caches and global development configurations. Developers began questioning how the data was being uploaded. In earlier versions, the feature was enabled by default, while the client did not provide a clear option to disable it.
Some developers reported that even after turning off privacy and indexing-related settings, background processes could continue attempting to upload the data package.
For enterprise developers, the issue went beyond product functionality and raised concerns over code assets and data security. After the incident became public, some companies reportedly stopped using the tool and began checking whether their code had been exposed. Zhipu responds with a series of changes
Within 72 hours of the incident, Zhipu introduced several changes to ZCode. On September 18, Zhipu issued an apology and acknowledged that the upload mechanism had not been adequately disclosed to users when the feature was initially launched.
On September 19, ZCode released version 3.14.0, removing the Repo Wiki entry as well as the relevant paths for generating and up repository snapshots.
On September 21, Zhipu open-sourced ZCode’s source code and released the results of its first round of third-party security checks. The company said the affected code data had not been retained and had never been used for model training.
In its latest announcement, Zhipu further confirmed that all data objects in the affected Alibaba Cloud OSS bucket, as well as the bucket itself, had been deleted. The deletion was verified by third-party organizations.
Zhipu also said ZCode would adopt a “no upload unless initiated by the user” approach going forward. Unless users actively initiate an upload, code, project files and other development data will remain on local devices and will not be automatically uploaded to the cloud.
The open-source version of ZCode has now been updated to version 3.14.3. Its source code is hosted on GitHub under the Apache-2.0 license.
Compensation for all users
Alongside the technical remediation, Zhipu also announced a compensation plan. All paid users will receive four weekly quota reset cards and four five-hour quota reset cards, each valid for one month. From September 28 to October 7, ZCode will also distribute 100,000 free Token packages to all users, with each package containing 100 million Tokens.
The controversy highlights a broader issue facing AI coding tools: as AI becomes increasingly integrated into software development, knowing where users’ code goes can be just as important as the capabilities of the underlying model.
For AI coding tool providers, there needs to be a clear distinction between data uploaded at a user’s request and data transferred automatically in the background. Data collection mechanisms also need to be transparent, with users given clear and verifiable controls over whether their data is uploaded. The ZCode incident may now be nearing its end, but as AI coding tools become more deeply embedded in enterprise development environments, how to balance AI capabilities, productivity and code security will remain a long-term issue for the industry.