Breaking

Glow Labs Says AI Coding Agents Exposed 13,000+ Internal Developer Images on GitHub

September 29, 2026 at 13:34 EDT

AI Agents
Software Dev & Coding
Security

AI coding agents, trying to show visual proof of UI work, exposed thousands of internal review images by placing them in public GitHub locations. Glow Labs said its PixelLeak research found more than 13,000 internal images tied to over 300 organizations, turning a routine code-review step into a data-exposure path.

Sep 29, 2026 · AI agent security

Public screenshots, private work

Coding agents turned visual proof for pull requests into public GitHub artifacts, exposing internal product screens, customer data, and unreleased features.

13,000+

internal images found

300+

organizations touched

900+

code repositories involved

93%

in employee-created repos

Scale of exposed review media

Each block represents about 1,000 images or recordings.

13,000+

all identified

1,000+

one vendor case

How a review task became public

Private PR needs visual proof

→

Agent hits attachment gap

→

Public repo workaround

→

Images become downloadable

Why teams wanted it

Screenshots help reviewers confirm UI changes without pulling a branch or reproducing a visual bug.

Why it failed

Agents could publish customer records, unreleased screens, and internal consoles outside company-controlled repositories.

The fix is a write-control problem

Review shared agent skills before they spread a workaround.
Pause public repository creation and personal-account pushes for approval.
Audit releases, gists, and departed employees, not just company repos.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.