LLM Guards

Essential security systems that filter malicious inputs and unsafe outputs (e.g., prompt injection, PII leakage) to keep LLMs aligned with policy.

LLM Guards (or Guardrails) are critical safety layers: they act before and after the core model to enforce strict ethical and operational guidelines. They actively mitigate key risks like prompt injection, data leakage (PII/secrets), and the generation of harmful content (illegal advice, hate speech). Frameworks such as Meta's Llama Guard or NVIDIA's NeMo Guardrails use advanced classifiers and scanners to ensure all model outputs remain safe, relevant, and compliant, preventing reputational damage and misuse in production environments.

https://docs.guardrailsai.com/