Everyone Is Buying AI Guardrails. But Agents Have the Keys to the Car.

The post argues that traditional AI security focused on prompts and outputs is insufficient for agents that act on systems; organizations must adopt "agentic security"—providing visibility into agents, governance of tools and identities, and runtime protection to monitor and control the full execution path from prompt to API/action to prevent unintended or harmful automated actions.