Guardrails

Guardrails are rules and filters that stop an AI tool from producing harmful, off-topic or confidential output.

In practice: They can block certain topics, keep a support bot to your products, remove personal data, check facts against your content or require human approval before an action. They sit around the model rather than inside it.

Why it matters when picking a tool

For customer-facing AI, ask what guardrails exist and whether you can configure them. Check what the tool does with sensitive data such as names, card numbers or health information, and whether it logs conversations for review.