As AI agents move from experimental playgrounds into core production workflows, the stakes for security have never been higher. When an agent acts on your behalf the system must answer one critical question: “Who authorized this, and which specific agent is doing the work?”