Agentic AI can reason, plan, and act—but it also creates new security risks that traditional governance models were never built to handle.