Comment by Go7hic

GitHub Agentic Workflows lack a trust boundary: attackers can inject instructions through public issues and trick the AI agent into leaking private repositories belonging to the same organization.