The PracticeThe PracticeThe PracticeOperating infrastructure

The Practice · Glossary

Guardrails

AI term; checked November 9, 2026

Guardrails are the rules and checks that keep an AI system inside set limits: what it may say, what it may do, what it must hand to a person and what it must refuse. They can be written into instructions, enforced by software around the model, or both.

Useful guardrails for an agent are concrete: never state a value as fact, never describe who a home suits, never send without approval. See The Guardrails Pack.

Educational, not professional advice. Nothing on this page is affiliated with or endorsed by its publisher.