Guardrails

In one sentenceRules and checks that keep AI output safe, on-brand and on-task.

What it means

Guardrails can be built in by the AI company (refusing harmful requests), added by a product (blocking off-topic answers), or set by you ("never mention pricing", "never promise results").

How to use it

• Write your own guardrails into your Custom Instructions or Persona: words you never use, claims you never make, topics you avoid.
• For customer-facing bots, test what happens when someone tries to get it off track.

Related terms

Human in the Loop Prompt Injection System Prompt Persona