Sort Controls by Where They Sit
Identify whether a proposed mitigation controls AI input, model behavior, tool authority, output, or monitoring.
Sort each mitigation by the layer where it primarily acts. Input or retrieval layer Model instruction layer Tool authority layer Output or monitoring layer Quote retrieved web pages as evidence and strip hidden metadata before model context. System prompt says external content cannot change the task or tool policy. CRM update tool requires a user-confirmed record ID and policy check. Alert when the model tries to reveal hidden instructions or change recipients. Retriever enforces the requesting user's document permissions. Developer message defines citation rules for regulated-policy answers. Refund tool rejects amounts above the agent's allowed threshold. Post-generation filter blocks private…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in