Classify common AI guardrails by the layer where they reduce risk.
Place each AI guardrail in the layer where it primarily reduces risk. Governance Input Model Output Action Monitoring Named risk owner and approved use-case scope Prompt redaction for account IDs and API keys System instruction that limits the assistant to policy-grounded answers JSON schema validation before a draft enters the ticket system Human approval before an AI agent updates customer billing Weekly review of blocked prompts and reviewer overrides Launch criteria requiring eval results and rollback owner Refusal when retrieved sources do not support the answer Tool allowlist that exposes only the duplicate-review endpoint
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in