Correct system-assigned attack verdicts in a completed scan report and recalculate
the Risk Score to reflect your team's validated findings.
| Where Can I Use This? | What Do I Need? |
|---|
- Prisma AIRS (AI Red Teaming)
|
- Prisma AIRS AI Red Teaming License
- Prisma AIRS AI Red Teaming Deployment Profile
|
When a scan completes, AI Red Teaming uses judge models to assign a verdict to each
attack automatically. You can review the judge's reasoning for each verdict, correct
individual verdicts, and trigger risk score re-evaluation to regenerate all report
metrics based on your team's validated findings. Every correction is recorded in an
audit trail that includes the judge's verdict or updated verdict (if manually
overridden), user identity, and timestamp
Your override verdict acts as feedback for the judge on subsequent scans that you run for the target.
With this feedback, the judge is less likely to flag the same kind of response again. The feedback applies across scan types and only to the target you scanned. It does not affect other targets. To give the judge context before you run the first scan, upload documents under
Target purpose and behavior when you
add or edit the target.
