Glossary
AI guardrails
Controls around an AI system that limit what it can see, say and do, checked in code, not just in the prompt.
AI guardrails are the technical and procedural controls that keep an AI system within safe, intended behavior. They include input and output filters, permission limits on tools, validation of actions, spend and step limits, and human approval for risky steps, applied around the model rather than trusted to the prompt alone.
A prompt that says "never reveal customer data" is a request, not a control. Guardrails are enforced outside the model, so they still hold when the model makes a mistake or is manipulated.
They work in layers: check what goes in, limit what the model can reach, check what comes out, and review what it did. Cloud providers and open-source tools offer filters, and the strongest controls are usually your own permission and approval rules.
Guardrails reduce risk and do not remove it. They need testing, which is where [AI red teaming](/glossary/ai-red-teaming) comes in.
Start in 60 seconds
Put this to work on your project.
Get an AI-assisted, architect-reviewed scope, cost range, and timeline. Free, no login.