Securing AI Agents — Prompt Injection, Least Privilege, and SandboxingIn security reviews for agents, I now start with one question: “If this agent…