F5 AI Solutions > AI Runtime: Building Secure and Optimized AI Apps with F5 Guardrails & Red Teams Source | Edit on
Lab 3 - Red Team – Campaign Exploration¶
Objective: Review results from a pre-run Red Team campaign and understand how to communicate findings.
Note: Red Team campaigns can take anywhere from 30 minutes to several hours tocomplete and incur model usage costs.
For this lab, we will not be launching any new attack campaigns. Instead, we’ve prepared a pre-run campaign so we can focus on reviewing and interpreting the results without waiting for execution or consuming additional resources.
Steps:
- Open the Red Team reports by clicking on the Reports in the
navigation menu on theleft.
Observe that All Attack Campaign Run reports are displayed. At a quick glance you will see the high level details including which campaign was run as well as the CASI score. Click into the ‘Full report’ report.
This will bring up an overview of the results. Let’s investigate the details. Click on ‘View Raw Data’ in the upper right handcorner.
Here you will see all the details of every attack prompt that was sent against the target. Let’s just filter on the successful attacks. Click on Filter, and select True.
Inspect one failed attack chain — note why it succeeded.
Reflection:
- Which attack category had the highest failure rate? Why do you think
- that is?
- Which findings would you escalate as “must-fix before production”
- vs.“track andmonitor”?