AI agents have mistakenly exposed 13,000 internal screenshots
More than 13,000 internal screenshots were discovered in the open on Github. These screenshots were posted by AI agents, on behalf of 343 different organizations (including those on the Fortune 500 list).
Glow Security recently discovered more than 13,000 internal screenshots publicly exposed on GitHub. These screenshots were uploaded by code generating agents, who interact with GitHub via CLI (command line interface) and can’t attach screenshots to issues that way, since GitHub only allows that through the browser. The agents did this by setting up a public repo (often in the dev’s personal namespace) where they could dump their screenshots, along with sensitive information like client info, passwords, and ideas for future development. Around a third of the impacted companies were using gitshot, an open source tool for taking screenshots that defaults to being public. Some of the agents also went so far as to install gitshot!
This is a tale of agent ingenuity and workarounds. When faced with a barrier, agents will find a way around it, often in less secure ways. As someone who works with agents and Git repos daily, I will definitely be keeping a closer eye on any public repos created by my agents that I haven’t explicitly requested. The problem here isn’t just that these companies had a data breach; it’s that they didn’t know it happened.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides

Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles