🛡️ Safeguards

By Zack Hill
3 min

Safeguards allow you to block AI outputs based on specific keywords in responses. They are perfect for preventing harmful, unethical, and dangerous responses from Await Cortex. They also help protect against malicious prompt injections and jailbreaking.

Feature: Safeguards

Instructions

  1. Go to the agent's tab and click on an agent you have created

    image-20240424-185813.png

  2. Once inside your agent panel you will have 4 tabs to choose from - Configuration, Chat Interface, Analytics, Cached Answers, and Safeguards. Click Safeguards.

image-20240425-021349.png

  1. Now you’ve arrived in the safeguards screen. This is where you can configure them.

    image-20240425-021726.png
    Safeguard details

     

    image-20240425-024133.png
    Safeguard rows can can be expanded

     

    image-20240425-024047.png
    Easily edit safeguards

     

  2. To create a new safeguard click the “Add New” button in the top right.

    image-20240425-024103.png
    Click “Add New” to create a new safeguard

     

  3. Now you are in the create safeguard screen - fill in the fields to create your first safeguard:

image-20240425-024008.png
Creating a new safeguard

Fallback and Warning Example

A Fallback makes the AI respond with a canned response based on your keyword flags

A Warning allows the AI to generate a response put provides a disclaimer for it’s answer

Safeguard Example: Medical Disclaimer

image-20240425-030418.png
Note the message and flag keyword highlighted

What it looks like in the agent:

Fallback

image-20240425-025804.png

Disclaimer

image-20240425-025754.png

Replace

Replace safeguards allow you replace - keyword flags with a canned message.

image-20240517-235555.png
New Update 5/17/24

 

Related content