What is an AI guardrail?
What is it?
A guardrail is a set of rules that controls what goes into or out of an AI assistant: block personal data, detect prompt injection, refuse off-topic requests.
When to use it
- Before putting an assistant in production, to limit leaks and abuse.
- To prove with test cases that a rule works (GDPR, AI Act).
Page fields
| Type | Personal data, prompt injection, toxic content, off topic, answer format. |
|---|---|
| Action | Block, mask, flag or rewrite. |
| Rules | What must be blocked or allowed, in plain words. |
| Test cases | Texts with the expected result: block or allow. |
| Code | Optional: the implementation, for example in Python. |
How to use it
Download the JSON: the rules instruct a checker model, and the test cases measure its mistakes.