A developer adds an input guardrail to a customer-facing agent. What role should that control play?

1. Guarantee that every future tool response is safe
2. Replace all tool-level authorization checks
3. Give the model permission to disable security controls
4. Check relevant input constraints as part of a layered safety design