AI Security
Search K

Appearance

Review a Red Team report ​

Overview ​

When you run a Red Team attack campaign, AI Security creates a report. In the UI, review the report by selecting Reports > the report of your choice. Use the report to see how well your model resisted attacks, identified vulnerabilities, and what to fix next.

A report includes:

  • A summary of scores and attack counts
  • Signature attack results, broken down by intent category and attack vector
  • Results mapped to selected OWASP LLM Top 10 2026 categories
  • Agentic resistance, agent attack prompt, and operational attack results
  • Suggested actions

Before you begin ​

Before you begin, make sure you have:

  • Access to the AI Security UI: If you use SaaS, the URL is typically https://us1.calypsoai.app.
  • Permission to view Red Team reports: For more information, see Permissions in AI Security.
  • At least one completed campaign run: To start a campaign, see Get started with AI Red Team.

Open a report ​

  1. Log in to the AI Security UI.
  2. Select AI Red Team > Reports.
  3. Select a report to review.

The report shows status and the associated campaign.

From the top of the report, you can also:

  • Select the print icon to print the report.
  • Select View raw data to see the underlying report data.
  • Select Retain data to keep the report data.

Review the summary cards ​

The summary cards provide:

CardWhat it shows
CASI scoreThe Comprehensive AI Security Index (CASI) score for signature attacks.
Agentic resistanceThe score for agentic resistance attacks.
Operational attacksResults for operational attacks.
Recommendations
Total attack countThe total number of attacks in the run.

Review signature attack results ​

The Signature attack results section shows results for the attack pack used, such as the October 2026 attack pack.

This section includes:

  • Vulnerabilities, by intent category: A chart of successful attacks for each intent category, such as Misinformation or Toxicity/hate.
  • Comparison of Attack success by attack vector vs. intent category.

Review attacks by framework ​

The Attacks by Framework: OWASP 2026 section maps attack success rates to categories in the OWASP Top 10 for LLM Applications 2026. This section covers the following categories:

  • LLM01: Prompt Injection
  • LLM07: Misinformation

Review agentic and operational results ​

This shows results for other attack types:

  • Agentic resistance results
  • Agent attack prompts results
  • Operational attacks results

In addition, you can View agentic fingerprints. Specifically, that allows you to analyze the logic by:

The agentic fingerprint allows you to review how AI Red Team analyzes and ends the attack, step-by-step.

References ​

For more information, see:

Updated at: