When an AI agent finds the weak API

AI agents turn weak permissions into operational risk. Use scoped access, action controls, audit trails, and rehearsed shutdown protocols to contain them.