332
Productivity & Workflow348
Automation & Workflow237
Software Development243
Marketing & Growth193
Writing & Content Creation193
AI Infrastructure & MLOps190
Photography & Imaging154
Design & Creative167
Data & Analytics161
Customer Support151
Education & Learning136
Voice & Speech133
Sales & Outreach129
Operations & Admin117
Researchers found Claude Code settings files shared on public GitHub repos, sometimes containing database passwords, API tokens, and private keys.
In short: Security researchers say a Claude Code settings file has been accidentally uploaded to public GitHub projects, exposing some passwords and access keys.
AI security researchers at Apiiro reported that Claude Code, a coding assistant from Anthropic, can save sensitive information into a local settings file called .claude/settings.local.json. This can happen after a user approves an automated command and picks “Yes, and don’t ask again.”
The problem is that the saved rule sometimes includes the full command text, and that command can contain secrets like database passwords, API tokens (keys that let software talk to other services), and private keys (like a master key for a locked door). The settings file is meant to stay on a person’s computer.
Anthropic’s documentation says this file is added to a developer’s global “ignore” list, which normally helps prevent it from being uploaded to GitHub. Apiiro says that protection does not always carry over to each project folder, so the file can be committed by accident, especially if a folder becomes a GitHub repository later or the project is moved to another machine.
Apiiro found over 100,000 cases where this settings file appeared in public GitHub repositories. About 1,100 files from 63 organizations contained real secrets, according to the researchers. Apiiro also warned that even if a file is removed, the secrets can live on in “forks” (copies other people made).
If a password or key is exposed, strangers may be able to get into company systems, run up cloud bills, or access private data. Anthropic staff said on Hacker News that the team is looking into ways to prevent this, and recommended using secret scanners, although Apiiro found some secrets were missed by common scanners.
Source: Arstechnica