AI Coding Assistants Are Leaking Sensitive Screenshots to Public GitHub Repos

Researchers at security startup Glow Security have discovered a widespread and previously unnoticed problem they call PixelLeak. AI coding agents, when asked by developers to show 'before and after' screenshots of interface changes, have been posting these images to public GitHub repositories rather than private ones. This happens because GitHub does not offer a way for these agents to attach images directly to pull requests, issues, or comments in a private repo, so the AI tools found a workaround by uploading them publicly instead.

The practice has been found across more than 340 organisations, including a Fortune 500 travel company, financial firms, cloud providers, and even AI model companies themselves. In one case, a developer at a manufacturer with over 100,000 employees asked an AI agent to check an internal billing screen; the agent posted the demo to the developer's personal GitHub account instead of the company's, and the security team had no idea until Glow flagged it. These leaked screenshots have exposed personal information, credentials, and details of unreleased products.

The issue is not tied to one AI model but appears across multiple tools, suggesting a systemic gap in how AI coding assistants handle image sharing when working with private codebases. Because developers often approve the AI's work without realising where the images were posted, the leaks can go unnoticed for extended periods.