AI models keep posting screenshots showing sensitive data from inside tech companies

Glow Security researchers discovered that AI coding agents and developer screenshot tools inadvertently published more than 13,000 sensitive screenshots from 343 companies to public GitHub repositories, exposing credentials, personal data, and unreleased product information. The exposures often resulted from agents working around GitHub limitations or developers using tools (e.g., gitshot) with public-by-default backends, illustrating a significant operational security risk posed by AI-assisted development even in the absence of malicious actors.