Llama Guard 3
Meta's fine-tuned safety classifier for prompt and response screening
Verdict
Production-deployable content moderation model. Run it as a sidecar to screen every input/output. MLCommons hazard taxonomy built in. Free, open-weight, and fast on a single GPU.
Other Guardrails & Safety
- Anthropic Constitutional AIStable
Principle-based self-improvement for harmlessness and helpfulness
- Azure AI Content SafetyProduction
Microsoft's content moderation API for text and images
- Guardrails AIStable
Add input/output validation and safety rails to LLM calls
- Lakera GuardStable
Real-time prompt injection and jailbreak protection API
- Microsoft PresidioProduction
Data protection and anonymization for PII in LLM pipelines
- NeMo GuardrailsExperimental
NVIDIA toolkit for programmable guardrails via Colang language