Why AI agents cannot be trusted to secure agentic AI yet

The idea of using AI agents to secure AI agents seems compelling, but in doing so, we risk unnecessarily expanding our attack surface.