# What are AI guardrails?

> Rules and checks that keep an AI support agent on topic, within your policies and away from actions it shouldn't take on its own.

URL: https://office.greatwork.company/glossary/ai-guardrails
Last updated: 2026-11-21

Rules and checks that keep an AI support agent on topic, within your policies and away from actions it shouldn't take on its own.

Guardrails can be written rules ("never promise a discount"), settings (answer only about this business), automatic handoffs on sensitive topics, rate limits, and confirmation steps before actions.

They work best in layers: plain-language rules for tone and policy, and hard limits in the software for anything that touches money or accounts.

## Where you'll use it

- [Guidance rules and tone](https://office.greatwork.company/assist/features/guidance-rules): Tell the assistant how to behave in plain sentences, pick a tone and answer length, and test any rule before customers see it.
- [Manage Stripe subscriptions in chat](https://office.greatwork.company/assist/features/stripe-actions): Connect Stripe and the assistant can show a verified customer their plan, renewal date and invoices, send a billing portal link, and cancel at period end after they confirm.

## Related terms

- [Prompt injection](https://office.greatwork.company/glossary/prompt-injection): Text written to trick an AI model into ignoring its instructions, typed by a user or hidden in content it reads.
- [AI hallucination](https://office.greatwork.company/glossary/ai-hallucination): When an AI model states something that sounds right but isn't supported by any source, like an invented return policy, feature or price.

---

Office by Great Work. Questions or corrections: hello@greatwork.company
