Guardrails

Technical and policy constraints that limit AI system behavior — preventing harmful outputs, enforcing scope, and maintaining compliance in production.