AI-generated "Policy as Code" can introduce silent security flaws. Learn why "almost correct" isn't enough for LLM-driven access control.