Lakera Red

Interpreting Results

After a Lakera Red scan completes, you’ll receive detailed results about vulnerabilities discovered in your GenAI application. This guide explains how to read and prioritize your results.

Results Overview

The scan results page shows:

  • Risk Score - Percentage of attacks that succeeded (0-100%)
  • Severity - Categorized based on risk score
  • Objectives Tested - Number of attack objectives run
  • Harmful / Safe Evaluations - Count of successful vs. defended attacks

Severity Levels

Severity is calculated from your risk score:

Risk ScoreSeverityInterpretation
≤25%LowMost attacks defended; minor hardening opportunities
26-50%MediumSome vulnerabilities present; remediation recommended
51-75%HighSignificant vulnerabilities; prioritize remediation
>75%CriticalSevere exposure; immediate action required

Viewing Results

By Risk Category

Group results by attack category (Security, Safety, Responsible, etc.) to see:

  • Risk score per category
  • Which categories have the most vulnerabilities
  • Overall distribution of issues

This view helps identify which types of attacks your application is most vulnerable to.

By Test

Group results by individual attack objective to see:

  • Each specific test that was run
  • Success/failure rate per objective
  • Which exact vulnerabilities were found

This view helps pinpoint specific issues to fix.

Understanding Individual Results

Click on any result to see the full details:

All Tests Tab

When an objective has multiple test runs, see all attempts and their outcomes.

Details Tab

Details about how the result was assessed:

FieldDescription
ResultWhether the attack achieved its objective
ExplanationWhy the evaluator determined success or failure

The exact exchange between Red and your application:

  • User messages - The attack prompts sent to your application
  • Assistant messages - Your application’s responses

Multi-turn attacks show the full conversation sequence.

Comparing Scans

Use the Compare feature to track security improvements over time:

2

Select two scans

Choose Scan A and Scan B to compare (cannot be the same scan).

3

Review comparison

See side-by-side analysis including:

  • Configuration Context - Differences in recon context
  • Scan Result Comparison - Table comparing results by objective
  • Risk Category Analysis - Charts showing risk changes by category

Compare scans to:

  • Verify remediations were effective
  • Track security posture over time
  • Compare different configurations or system prompts

Providing Feedback

You can provide feedback on individual results to help improve Red’s accuracy:

  • Mark results as Good (accurate assessment) or Bad (incorrect assessment)

This feedback helps Lakera continuously improve evaluation accuracy.

Exporting Results

Export your results for reporting or integration with other tools:

JSON Export

Full scan results including:

  • All conversations and responses
  • Evaluation details and scores
  • Objective IDs and metadata

CSV Export

Flattened format with columns:

  • Objective name
  • Explanation
  • Conversation (formatted)
  • Error messages (if any)

Next Steps