A security investigation by Glow Security found that AI coding agents uploaded more than 13,000 internal screenshots belonging to hundreds of organizations to public GitHub repositories, a phenomenon the researchers named PixelLeak. Leaked material included credentials, control panels, personal information and details of unreleased products, with no attacker involved, but rather the agents' lack of confidentiality boundaries and permission controls. Recommendations included reviewing employee accounts and curtailing the use of self-running tools.