Skip to main content

Framework mapping

Every attack objective and finding is mapped to industry standards, so you can report results in the frameworks your auditors and risk teams already use.

Where mappings appear​

  • Run details: the Standards & Framework Coverage cards and the Categories table, which you can switch between standards. See Run details.
  • Full report: the OWASP and MITRE ATLAS coverage panels, and the mitigations and citations on each finding. See Full report.

Frameworks​

FrameworkCoversMeasured in
MITRE ATLASAI threat techniquesTechniques
NIST AI RMFAI risk managementSubcategories
OWASPLLM and agentic application risksCategories
ISO/IEC 42001AI management systemsControls
EU AI ActRegulatory requirementsArticles
CIS Controls v8Critical security controlsSafeguards
ISO/IEC 27001Information securityControls
SOC 2Trust Services CriteriaCriteria
CWECommon software weaknessesCategories (in the Categories table)

On each framework card, the count, such as 8 / 19 techniques tested, shows how much of the framework the run reached. A Quick scan reaches less than the Full library.

OWASP categories​

OWASP categories come from two lists: LLM IDs from the Top 10 for LLM Applications (2025), and ASI IDs from the Top 10 for Agentic Applications.

IDCategory
LLM01Prompt Injection
LLM02Sensitive Information Disclosure
LLM04Data and Model Poisoning
LLM05Improper Output Handling
LLM06Excessive Agency
LLM07System Prompt Leakage
LLM08Vector and Embedding Weaknesses
LLM09Misinformation
LLM10Unbounded Consumption
ASI01Agent Goal Hijack
ASI02Tool Misuse & Exploitation
ASI03Identity & Privilege Abuse
ASI05Unexpected Code Execution
ASI06Memory & Context Poisoning

LLM03 (Supply Chain) is not covered by the attack library.

Mappings on a finding​

MappingExample
OWASP category and mitigationLLM09, with guidance for that category
MITRE ATLAS techniqueAML.T0063 Discover AI Model Outputs
MITRE ATLAS mitigationsAML.M0002 Passive AI Output Obfuscation. Some techniques have no published mitigation, and the finding says so.
NIST AI RMF subcategoriesMEASURE 2.7, MANAGE 4.1

One attack can map to several categories and frameworks, so category totals can add up to more than the number of attacks.

On this page