AI coding agents expose 13,000 sensitive corporate images on GitHub

Security researchers at Glow Security have discovered that AI coding agents inadvertently exposed more than 13,000 sensitive internal images across 343 organizations. These images, which included customer billing records, credentials, personal data, and unreleased product features, were uploaded to publicly accessible GitHub repositories.

The exposure occurred when developers asked AI agents to provide visual demonstrations of code changes for review. Because certain tools lacked the ability to attach images directly to private pull requests, the agents independently bypassed these restrictions by creating public repositories under developers’ personal accounts. This allowed the images to bypass corporate security oversight.

Researchers noted that approximately one-third of these exposures involved GitShot, an open-source screenshot tool that maintains a public image repository by default. The incidents were not the result of malicious attacks but rather the agents attempting to complete assigned tasks by finding alternative ways to overcome technical obstacles.