WitnessAI Documentation
List
Start Here
Start Here
Analytics
Analytics
Conversations
Conversations
Users
Users
Alerts
Alerts
Policy
Policy
Administration
Administration
Apps & Resources
Apps & Resources
Release Notes
Release Notes
Administrator Guide
Administrator Guide
Guardrails
Guardrails are the checks a policy runs on prompts and responses. Each guardrail looks for one kind of risk and takes the action you set, such as Allow, Warn or Block. You add and configure guardrails on a policy, in its Guardrails section.
Available guardrails
Guardrail | What it does |
Identifies what the user is trying to do, and applies Allow, Warn, Block or Route. | |
Detects sensitive data and applies the controls you set. | |
Checks model responses for harmful content, and applies Allow, Warn or Block. | |
Checks identity instructions and keeps responses consistent with them, and applies Allow, Warn, Block or Route. | |
Analyzes prompts to protect the model, and applies Allow, Warn or Block. | |
Looks for patterns across several activities within a time window. | |
Rates the risk of a prompt, and applies Allow, Warn or Block. |
Configure a guardrail
Open a policy, select Guardrails, and choose the guardrail you want to configure. Every guardrail is set up from this same list.
To build a custom guardrail of your own, see Guardrail Studio.
Made with Bullet