Research has revealed that the AI coding agent Claude Code is inadvertently exposing sensitive credentials belonging to some developers through public GitHub repositories. More than 100,000 Claude Code local settings files have been uploaded to public repositories, involving over 2,500 live credentials across nearly 2,000 code repositories and encompassing confidential data from 63 organizations of various types. The root cause lies in how Claude Code handles authorization for automatically executed commands: when users select the "don't ask again" authorization rule along with the associated command content, these are written into a local settings file. This file should be added to the global .gitignore to prevent it from being committed, but under certain circumstances it may still be uploaded by mistake—and Claude Code sometimes even commits the file on its own. Even if the leaked file is deleted from the original repository, generated branches may still retain the keys. Previously, related bug reports were closed due to prolonged inactivity. The Claude Code team is currently investigating preventive measures, including a possible built-in secret-scanning feature, and recommends that users employ secret-scanning tools to clean up sensitive information.