TEGO AI has reported a critical security flaw in Anthropic's Claude Code tool, allowing unauthorized access to files located outside the project repository. The issue arises when a user clones a repository that contains a symbolic link, enabling Claude Code to send sensitive files in its first request without user awareness. This flaw builds on previous vulnerabilities and shows that the security checks are inadequate since they only monitor link names within the repository instead of the final file paths.
Tego AI emphasizes the importance of understanding security boundaries when using AI coding tools and suggests that the existing model could lead to significant risks as enterprises adopt these technologies.