Skip to content
GitHub Discord

Review audit results

Open a completed audit from Audit history to understand the risks found in your agent and decide what to fix first.

The AI Trust level summarizes your agent’s behavior during the audit with a grade from A to D:

GradeLabelMeaning
AExcellentNo issues were found in the tested behavior.
BGoodNo high-severity issues were found; improvements are recommended.
CInsufficientExploitable attacks were identified and need remediation before production.
DPoorFailures were found that undermine trust in the agent.

Review the grade alongside the severity counts and overall assessment to prioritize your investigation. For a Light audit, check how many scenarios were covered, since the untested ones still need a Full audit. The OWASP coverage section groups findings that have OWASP category mappings.

Audit results with trust grade, severity counts, scenario coverage, and assessment

The Risk model tab lists the risk scenarios identified for your agent. Expand a scenario to understand what could go wrong, which assets are at risk, and why the scenario matters. Linked findings show the issues discovered while testing it.

For a Light audit, the table separates covered scenarios from those left uncovered, which only a Full audit will test.

Expanded risk scenario with affected assets, rationale, and a linked finding

Open the Findings tab and expand a finding to inspect:

  • The issue description and severity.
  • Its likelihood, impact, and taxonomy categories, which can include one or more OWASP categories.
  • Attack examples linking to recorded test results.
  • Related risk scenarios and recommendations.

Follow an attack example to inspect the agent’s inputs and responses. Open Recommendations to review the proposed fixes and their priorities. Each recommendation links back to the findings it addresses.

Expanded finding with attack examples and linked recommendations

When the PDF report is ready, click Download Report to save a copy and share the assessment with your team.

Download Report button on a completed audit

Audit report contents, including methodology, risk model, findings, and remediation roadmap

Use the resources in Audit outputs to investigate findings and test your fixes:

  • Audit dataset contains generated scenarios that you can review, refine, and reuse in evaluations.
  • Evidence findings links to an evaluation containing the results that support the findings.

Audit dataset and evidence evaluation with scenario and result counts

Review the generated scenarios before using them as regression tests. After improving your agent, run an evaluation against the audit dataset and compare results to check whether the fixes worked.

If no issues were found, there may be no evidence evaluation. Any problems saving the results are shown in this section.

For a more thorough assessment, click Run Full Audit from a completed Light audit. This launches a separate Full audit using the same agent and input evaluations. Both runs remain accessible from Audit history.

Run full audit button on a completed Light audit