Chinese AI Firm Z.ai Disables Coding Assistant Features After Security Concerns

Chinese artificial intelligence company Z.ai has disabled some features of its AI-powered coding assistant after users raised concerns that the tool was sending local code repositories to overseas cloud servers without their knowledge.

The issue reportedly involved Z.ai’s ZCode coding assistant and its “Codebase Indexing” feature. The feature, which was enabled by default, allowed the system to analyse a user’s local codebase to provide more relevant assistance. However, security researchers and developers found that information from local repositories could be transmitted to remote servers, raising concerns about the handling of sensitive source code.

Z.ai acknowledged the problem and apologised to users. The company said it had identified and fixed the vulnerability and temporarily disabled the affected functionality while it reviewed the system.

The incident has raised broader questions about the security risks associated with AI coding tools. Such services often require access to developers’ files, repositories and software projects, which can contain proprietary information, credentials or other sensitive material. Unintended transmission of such data can therefore create significant privacy and cybersecurity concerns.

An independent security assessment also examined the behaviour of the ZCode assistant. Developers reportedly questioned the company over how information was collected, where it was stored and whether users could ensure that transmitted data was permanently deleted.

Z.ai has said it would strengthen its security and data-handling practices following the incident. The company is also expected to introduce additional safeguards relating to data retention and user privacy.

The episode highlights the challenges facing AI companies as coding assistants become increasingly integrated into software development. While these tools can access and analyse large amounts of code to improve their responses, granting them such access also creates potential security risks. For developers and organisations, the incident underscores the importance of understanding an AI tool’s default settings, data policies and permissions before allowing it access to private repositories.