Glow Labs reveals AI coding agents have uploaded over 13,000 internal screenshots to public GitHub repositories. The research, called "PixelLeak," discovered images originated from developers at over 300 organizations. Each case began when a developer asked an AI agent to prove a visual change worked, such as a user interface layout fix.

Public Repository Screenshot Exposure (Image Source: Glow) Around a third of affected organizations had developers using gitshot, a small open-source tool that publishes screenshots for code reviews. Over 100 public accounts leaked internal work, including a major frontier AI company, a financial services firm with screenshots of an internal treasury and settlement console, and a dollar withdrawal screen for a named institutional client alongside two screen recordings that walked through the money-movement console.

Within a week, over a dozen agents had encoded the practice as a reusable skill, uploading more than a thousand screenshots and screen recordings, along with summaries of features weeks or months from release. To enhance agent security, organizations should monitor for shadow AI agents, avoid blanket auto-approval of agent actions, review shared agent skill and instruction files, and remove untested packages such as gitshot. Glow suggests pre-execution runtime hooks to block or approve agent actions: creating new public repositories, pushing to personal accounts, pushing to gists, and switching a repository from private to public.