Safety & Alignment
Guardrails
Programmatic validators that constrain LLM inputs and outputs against policies.
Definition
Guardrails frameworks (Guardrails AI, NeMo Guardrails, Llama Guard) intercept model I/O to enforce schemas, detect PII, block jailbreak attempts and ensure topical relevance. They complement provider-side safety with application-specific rules.
Common use cases
- Trust & safety
- Compliance
- Domain restriction