Views:

Framework scans score your AI Scanner results using the recognized industry security standard to see how ready your AI application is for production.

When you run a scan with a framework attack preset, AI Scanner maps each attack to the framework requirements it tests and reports the results as a readiness score for each level or category the framework defines.
A framework scan is a standard AI Scanner scan that runs a framework attack preset. The preset selects the attack objectives, techniques, and modifiers required to test the framework's requirements, so you do not have to pick them yourself. After the scan completes, AI Scanner maps each finding to the requirements it tests and rolls the outcomes up into a readiness score.
Use a framework scan when you want a scored, requirements-based view of your security posture instead of a list of individual attack results.
This topic covers:

How framework scans work

Important
Important
Select the framework's attack preset at scan setup for full coverage of its requirements. AI Scanner evaluates framework results on every scan. However, without the preset, the attacks that test each requirement may not run, so requirements can end up not assessed or only partially assessed.
AI Scanner supports the following frameworks:
Framework
Standard
Scored by
OWASP AISVS
AI Security Verification Standard (AISVS)
Assurance level
The attack_preset setting accepts three values: OWASP, MITRE, and AISVS. Only AISVS is scored as a framework scan. The OWASP Top 10 for LLM Applications and MITRE ATLAS presets are coverage presets whose results appear as framework compliance mappings on every scan, with or without a preset. For more information, refer to AI Scanner scan results.
A scan uses one attack preset at a time.

OWASP AISVS

The Open Worldwide Application Security Project (OWASP) AI Security Verification Standard (AISVS) is a checklist of security requirements for AI systems, grouped into three ascending assurance levels. AI Scanner assesses the AISVS requirements that attack testing can verify, covering areas such as prompt injection defenses, input validation, model behavior, output safety, memory, orchestration, and adversarial robustness, and reports the result as a readiness score for each assurance level.
Level
Intended for
Level 1
Baseline protections every AI application should have
Level 2
Defenses expected of applications that handle sensitive data
Level 3
The most rigorous controls for high-risk or critical systems
The AISVS score reflects only the requirements that can be evaluated through attack testing, not full AISVS certification. AISVS also defines controls that attack testing cannot observe, such as training data handling and infrastructure hardening.
To run an AISVS-scored scan, refer to Run an AISVS scan.