Starting a red teaming run
Red teaming runs let you run automated red teaming against a model.
To start a red teaming run, you need to choose a list of behaviors to search for.
With a list of behaviors, we can start a red teaming run with the following steps:
-
Login to the Haize Labs Dashboard and select the Red Teaming tab on the left sidebar.
-
On the Red Teaming page, press the ‘Create Evaluation’ button.
-
Configure your red teaming run to search for the content you want.
-
Give a name to the red teaming run (this is just for you to see).
-
Select the Red Teaming evaluation type. You can also run fuzzing and data replay evaluations. For now we’ll stick with red teaming.
-
Select the model you want to test. Models are AI systems API endpoints you have set up in the Haize Labs platform. See Creating a Model for more details..
-
Input the behaviors you want to search for. Behaviors are the content that the red teaming run looks to find.
-
-
Press the ‘Submit’ button.
Your automated red teaming run is now created! Your run will begin and results will be logged in the dashboard.
Read calling a detector to learn how to use your detector to catch unwanted content.