Glossary
Guardrails
What are AI guardrails?
Guardrails are the automatic checks and hard limits around an AI agent that stop it from doing what it must never do, such as sending a message, moving money or editing a system of record, whatever the model decides.
Also called: AI guardrails, Agent guardrails
Why it matters
A rule written only in an agent’s instructions can fail when the model is wrong or is manipulated by something it reads. Limits enforced by permissions and code hold either way, which is why an agent’s PRD lists every “never” with how it is enforced.
Example
A revenue review agent gets read-only access to the forecast workbook, so it can’t edit the forecast even if an email tells it to.