Home › Glossary › Safety › Guardrails

Intermediate · Safety

Guardrails

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. Runtime checks that constrain model inputs and outputs to stay within safe, useful bounds.

Technical Definition

Runtime checks that constrain model inputs and outputs to stay within safe, useful bounds.

How it works

Guardrails block disallowed content, validate output format (JSON schema), redact PII, detect prompt injection, and enforce business rules. Implemented as pre/post filters, classifiers, or systems like NeMo Guardrails and Llama Guard. Essential for any production LLM application.

Related Concepts

  • Prompt Engineering — The art of crafting effective input instructions to guide LLM behavior without changing model weights.
  • Safety — Designing AI systems to behave reliably and avoid causing harm under realistic conditions.
  • Red Teaming — Systematically probing an AI system for failures, exploits, and harmful behavior before deployment.