Researcher demonstrates GitHub's AI agent can be tricked into exposing private repositories

A security researcher found that GitHub's AI-powered agent can be manipulated through prompt injection to leak access to private repositories. The vulnerability highlights real risks in deploying AI agents with repository access—a timely concern as more indie developers integrate AI tools into their workflows.