Custom guardrail types
This table shows the custom guardrail types:
| Guardrail type | Description |
|---|---|
| Ban List | A filter that blocks the model from processing or generating content that contains prohibited terms. |
| Restrict To Topic | A boundary-enforcement mechanism restricts model responses to specified topic domains and rejects discussion of any prohibited topic. |
| PII Detection | A selective-detection system identifies and protects categories of personally identifiable information, such as Social Security numbers, credit card numbers, email addresses, phone numbers, physical addresses, medical IDs, and authentication tokens. The system applies configurable redaction or blocking based on the sensitivity level. |
| Policy-based |
An administrator specifies a policy in plain English. Example policies include: "Responses must not include financial advice" and "Do not discuss competitor products." During policy enforcement, the system sends the response to a separate large language model (LLM) call. The LLM evaluates the response against the specified policy. |