AI guardrails
Controls that keep an AI system inside its authorized and useful operating boundary.
Why it matters
A guardrail makes the intended boundary explicit. It reduces the chance that an ambiguous prompt, unexpected data, or model error becomes an unreviewed business action.
Use it well
- Define what the system must never do.
- Put permissions in the application layer.
- Give people a clear way to review, correct, and recover.