Urgent.News

What's breaking now, across thousands of outlets.

AI

PixelLeak: AI coding agents leaked 13,000+ internal screenshots to public GitHub repos

Security firm Glow disclosed on September 29, 2026 that it had found more than 13,000 internal screenshots and recordings, from developers at 300+ organizations, sitting in public GitHub repositories — downloadable by anyone, with no login, and invisible to the organizations' own security scanning because the repos lived under developers' personal GitHub accounts, not their employer's. What the…

In September 2026, security firm Glow disclosed the discovery of over 13,000 internal screenshots and recordings from developers at more than 300 organizations, inadvertently left in public GitHub repositories. These repositories, which were hosted under developers' personal accounts rather than their employers' corporate accounts, were accessible to anyone, anywhere, and went unnoticed by the organizations' own security scanners.

This issue arose due to GitHub's command-line interface (CLI) not having a way to attach images to pull requests, a feature only available through the browser at the time. To work around this limitation, coding agents, when instructed to show a UI change, would create screenshots using a workaround known as gitshot. This open-source skill, installed in 40+ coding agents, would default to creating a public repository under the developer's personal account and store the screenshots as public release assets, without the option to use a private or organization-owned repository.

The use of gitshot was widespread, with about a third of the affected organizations employing it. Glow also noted that the pattern of this behavior spread socially, with one engineer's agent initiating the practice, and within a week, over a dozen developers had adopted it as their own reusable skill, subsequently uploading 1,000+ screenshots and recordings, including unreleased features.

Confirmed examples of exposed data included a utility company's billing records under a manufacturer's developer account, and a financial services firm's internal treasury/settlement console, a withdrawal screen naming a specific client, and recordings of money-movement workflows. Despite GitHub's September 1, 2026 update to its CLI adding an --attach flag to address the workaround, this update does not retroactively secure previously exposed data, nor does it function on GitHub Enterprise Server.

Glow's report does not confirm whether the exposed images were downloaded by anyone other than Glow's researchers, which AI models were involved in real-world cases, or whether every found gitshot repository was indeed agent output. Glow's statement that other organizations are likely affected is an expectation rather than a confirmed count.

It's worth noting that Glow sells agent-control software, which might influence the framing of the finding. The incident has been rated as critical (10/10) due to the scope (a skill usable across 40+ agents, affecting over 300 organizations) and impact (customer billing data, an internal treasury console), making it a significant concern for any team that has ever needed to show a screenshot in a pull request.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

MCP structuredContent: Claude Code Drops Your Text Blocks

If you build an MCP server, the instructions you write into the text of a tool result may never reach the model. MCP (Model Context Protocol) lets an AI app such as Claude Code call tools on a server…

  • Claude Code interacts with MCP to request tool results.
  • Text blocks dropped, JSON data used instead.
  • Note field added to structuredContent to preserve important data.

The China Shock in the AI Era

The dominance of Chinese manufacturing and the rise of AI could create new economic imbalances and strains on a massive scale warns Yasheng Huang.

  • 70% of voters support AI restrictions in 10 US states
  • 52% of Americans more concerned than excited about AI
  • "China Shock" and "AI Productivity Shock" may create economic imbalance

More from Saturday 14 November →