Claude Code setting exposes passwords and API keys on GitHub

Anthropic's Claude Code tool is leaking user passwords and API keys to public GitHub repositories, according to reports from CyberSecurityNews, The Hacker News, and Cyber Press. The data exposure occurs when developers select the permission option labeled yes, and do not ask again. The AI coding assistant saves the executed command word-for-word in project setting files, which developers then push to public code repositories. [1][2][4][10]

Outlets differ on the exact scale of the exposed data across GitHub repositories. The Hacker News reported finding 1,122 public copies of the affected setting file containing secrets. GBHackers News and The 420.in both report that scans uncovered 2,539 exposed secrets spread across 1,994 public GitHub repositories. [2][5][7]

A password is a string of characters used as secret data to verify a user's identity. Anthropic offers Claude Code as a terminal coding agent for AI-assisted software development. If developers include plaintext credentials in commands approved through the persistent permission setting, the utility logs those credentials directly into the project configuration. [2][8][10]

Separate research from endpoint security company Forever Security highlighted additional risks involving Claude features, according to The Hacker News. Researchers demonstrated that a malicious browser extension executing scripts on claude.ai could leverage the Dispatch feature to trigger remote code execution on a user's computer via Claude Desktop. [3]