AI Agents Leak Over 13,000 Screenshots on Public GitHub, Exposing 343 Companies

Cybersecurity firm Glow Security recently found over 13,000 screenshots in public GitHub repositories created by AI agents, some containing sensitive data from tech companies. Researchers call this leak method "PixelLeak."

When humans ask AI agents to show before-and-after comparisons of project changes, the agents take screenshots. Due to GitHub's image hosting limits, agents may upload PNGs to new public repos to render them, accidentally exposing information.

So far, 343 companies have leaked sensitive data this way, including a major global tech company, a leading AI lab, a large enterprise software vendor, and a Fortune 500 travel company. Glow Security advises companies to check their data exposure and tighten AI permissions.