Skip to content
Glossary

Guardrails, explained

Guardrails are rules and controls that limit what an AI system can say or do, such as off-limits topics, data it must not reveal and actions that require approval.

They work in layers: in the agent's instructions, in filters that check input and output, and in technical permissions on connected systems.

A good guardrail has a clear way out. When the agent hits a limit, it declines politely or hands the conversation to a person.

Limits that exist only in the instruction text can be bypassed. The most important ones should be enforced outside the model.

Conversations aren't support.They're infrastructure.

Step into the new era of the relationship between companies and customers.

Create agent now