Security · 2h ago
GitHub AI Agent Tricked into Leaking Private Repos
Security researchers at Noma exploited GitHub's AI agent to extract contents from private repositories. The attack, dubbed GitLost, used prompt injection to bypass access controls. GitHub has since patched the vulnerability.
Meridian48 take
This highlights the ongoing challenge of securing AI-powered tools against prompt injection, a threat that remains underappreciated in enterprise settings.
Read the full reporting
GitLost: We Tricked GitHub's AI Agent into Leaking Private Repos →
Hacker News
prompt-injectiongithub-ai