Seeing AI Agents Is Not Enough — Security Teams Must Enforce What They Can Do

The article warns that discovering AI agents is only a first step and that enforcing least privilege across heterogeneous platforms requires intent-based controls, contextual access rules, and platform-agnostic identity governance; it highlights fragmentation in enforcement, emerging standards (OWASP, NIST, Australian guidance), and cites the ChatGPT AgentForger vulnerability as a real-world example demonstrating the risks of agent creation flows and identity inheritance.