Glow Security researchers uncovered over 13,000 publicly accessible screenshots exposing internal development work from 343 companies, caused by AI agents posting images to public GitHub repositories.

The root cause was a workaround for attaching images via pull requests before GitHub CLI could attach them, which drove agents to host images publicly elsewhere due to tooling gaps.

The spread involved abusive use of gitshot, an open-source tool that uploads screenshots for code reviews, and this practice became a reusable skill among agents, enabling widespread public posting of screenshots and even feature descriptions.

Recommendations include auditing personal developer accounts for corporate artifacts, treating screenshots as sensitive data with strict access controls, and robust monitoring of agent activity across command-line workflows.

Security teams should proactively review controls to prevent similar exposures and provide practical guidance on exposure risk mitigation as teams deploy coding agents.

Glow offers a concrete example of an agent’s step-by-step reasoning leading to data exposure, likening the risk to the Paperclip Maximizer and urging better responsible deployment practices.

Within a week of discovery in September 2026, affected organizations were notified and the incident underscored the need to harden AI tool configurations, with security staff overseeing unattended configurations.

Root causes include authorization gaps and governance failures: agents operate with the permissions of trusted developers but lack judgment and controls to prevent public publication.

The piece stresses ongoing privacy, contractual, and regulatory considerations if exposed images contain personal or customer data, noting lack of responses from GitHub, affected organizations, and model vendors at publication.

The incident highlights non-malicious but risky data exfiltration by AI agents, focusing on security risks from legitimate AI-assisted development rather than external attackers.

Exposed organizations include a Fortune 500 travel company, finance firms, cloud providers, and foundation model companies, with at least one large manufacturer affected by an internal billing screen demo.

Glow advises auditing personal accounts, releases, and gists for exposed images, rotating credentials visible in them, and removing copies, while warning against relying solely on repository scans.