πŸ›‘οΈ AIVSS Maturity and Prioritization

AIVSS risk factor evaluation maturity with version agnostic CVSS prioritization

System Management

πŸ“š AIVSS Methodology Guide

Overview

This tool implements the AIVSS-SSVC methodology described in the accompanying research paper.

πŸ”„ Workflow Overview

  1. Create AI Systems: On the Dashboard, define each agentic AI deployment in your organization
  2. Assess AIVSS Factors: Click the Assess button on an AI system to evaluate 10 AI-specific risk factors against OWASP Agentic AI risks
  3. Manage Vulnerabilities: Use the Vulnerabilities tab to create a library of IT/OT vulnerabilities with CVSS scores
  4. Link & Map: In the Dashboard, click a System's Link Vuln button to connect vulnerabilities to affected AI systems and specific risk factors
  5. Prioritize Risks: Use the Prioritization tab to view automatically calculated priorities based on AIVSS status Γ— CVSS severity

πŸ“‹ AIVSS Maturity Model

The AIVSS Multifactor Maturity Model (more like an assessment coverage than a true maturity model) tracks the depth to which each AI system has been evaluated across 10 risk factors:

Four Status Levels

Present (Mature): System has been evaluated. Results are conclusive that one or more risk factors are present.
None (Mature): System has been evaluated. Results are conclusive that none of the factors in the group are present.
Unsure (Immature): System has been evaluated, but results are inconclusive.
Not Evaluated (Immature): System has not been evaluated against this factor.

Parent-Child Factor Logic

Some factors (like "Tool Misuse") have subfactors. The parent factor automatically inherits the lowest maturity status of its children. This ensures that any immature subfactor assessment makes the parent factor immature as well.

🎯 Risk Prioritization Algorithm

The prioritization matrix follows the AIVSS-SSVC methodology (Appendix A, Table 7):

Step 1: Enumerate

Create all possible combinations of linked vulnerabilities Γ— affected AIVSS factors.

Step 2: Filter

Remove combinations where status = "None" OR CVSS severity is "None". Low severity vulnerabilities remain in the prioritization.

Step 3: Sort

Sort the remaining items by:

  1. AIVSS Status (descending): Present > Unsure > Not Evaluated
  2. CVSS Severity (descending): Critical > High > Medium

Step 4: Number Sequentially

Assign priority numbers (1, 2, 3...) based on the sort sequence. Lower numbers indicate higher urgency. Priority captures the sequence of the resulting sortβ€”nothing more. This represents a relative risk prioritization useful for assigning mitigation work and allocating resources.

πŸ”’ CVSS Integration

This tool uses a third party cvss4 JS library to calculate CVSS Base Scores from vector strings. This module does not support CVSS 2.0, but this could be added as the methodology is agnostic to the CVSS version. Supported versions:

  • CVSS 3.0 & 3.1: Industry standard vulnerability scoring
  • CVSS 4.0: Latest specification with enhanced metrics

Important: CVSS scores are ordinal values (ranked labels), not real numbers. The distance between 2.0 and 5.0 is not the same as 4.0 to 7.0. This is why the prioritization uses sorting and binning rather than arithmetic operations.

πŸ’‘ Best Practices

  • Start Simple: Begin with 1-2 critical AI systems before expanding to your full portfolio
  • Evaluate Systematically: Work through all 10 AIVSS factors for each system methodically
  • Document Evidence: Keep external notes on why each factor was marked Present/None/Unsure
  • Regular Updates: Re-evaluate as your AI systems evolve or new vulnerabilities emerge
  • Cross-Functional: Involve AI engineers, security teams, and risk management in assessments
  • Prioritize Action: Focus remediation efforts on Priority 1-3 risks first

πŸ“– Further Reading

For complete technical details, consult the AIVSS-SSVC methodology paper and Appendix A. This tool implements the multifactor maturity model and CVSS integration framework described in that research.

Global Vulnerability Library

πŸ”’ About the Vulnerability Library

Purpose: Maintain a centralized repository of IT/OT vulnerabilities (CVEs) that impact your agentic AI systems. Each vulnerability is scored using CVSS (Common Vulnerability Scoring System) to establish traditional security severity. The methodology is agnostic to the CVSS version and you can mix versions.

Workflow:

  1. Add Vulnerabilities: Create entries for relevant CVEs from your vulnerability scanner or threat intelligence feeds
  2. Input CVSS Vector: Paste the complete CVSS vector string (v3.0, v3.1, or v4.0 format)
  3. Auto-Calculate Score: The system uses a third party cvss4 library to calculate the Base Score and severity rating
  4. Link to Systems: From the Dashboard, use "πŸ”— Link Vuln" to connect vulnerabilities to affected AI systems
    • From there you can specify which AIVSS risk factors are impacted by each vulnerability
CVSS Severity Ratings:
Critical (9.0-10.0) High (7.0-8.9) Medium (4.0-6.9) Low (0.1-3.9) None (0.0)

πŸ’‘ Note: CVSS scores are ordinal values (ranked labels), not real numbers. The system uses proper ordinal sorting and binning as described in the AIVSS-SSVC methodology to avoid invalid arithmetic operations.

Risk Prioritization Matrix

🎯 How Risk Prioritization Works

Purpose: Automatically calculate relative risk priorities by combining AIVSS factor assessments (AI-specific risks) with CVSS vulnerability severity (IT/OT security risks). This implements the AIVSS-SSVC methodology described in Appendix A, Table 7.

The Algorithm (4 Steps):
  1. Enumerate: Create all vulnerability Γ— factor combinations
  2. Filter: Remove "None" status OR "None" severity (Low severity is kept)
  3. Sort: By AIVSS status (Present > Unsure > Not Evaluated), then CVSS severity (Critical > High > Medium)
  4. Number: Assign sequential priorities (1, 2, 3...) based on sort order, plus categories (A-I) based on status Γ— severity intersection
Two Viewing Modes:
  • Per System: View priorities for a single AI system in isolation. Use this to focus remediation efforts on one system.
  • Across All Systems: Global enterprise view that re-sorts and re-numbers priorities across your entire AI portfolio. Use this to identify your most critical risks organization-wide.

πŸ’‘ Key Insight: Priority numbers (1, 2, 3...) are ordinal ranks from the sortβ€”focus on lowest numbers first. Category letters (A-I) group risks by status Γ— severity type. See legend below for category definitions.

πŸ“Š Category Legend (A-I)

Categories represent the intersection of AIVSS Status Γ— CVSS Severity. They provide a quick visual reference for the type of risk, complementing the sequential priority ranking. Categories are organized by status, with color intensity indicating severity.

Present Status (A-C)

A Critical Concern

Status: Present
CVSS: Critical or High
Confirmed risk, high severity

B High Concern

Status: Present
CVSS: Medium
Confirmed risk, moderate severity

C Moderate Concern

Status: Present
CVSS: Low
Confirmed risk, low severity

Not Evaluated Status (D-F)

D Needs Evaluation - High

Status: Not Evaluated
CVSS: Critical or High
Unevaluated, high severity vuln

E Needs Evaluation - Medium

Status: Not Evaluated
CVSS: Medium
Unevaluated, moderate severity vuln

F Needs Evaluation - Low

Status: Not Evaluated
CVSS: Low
Unevaluated, low severity vuln

Unsure Status (G-I)

G Inconclusive - High

Status: Unsure
CVSS: Critical or High
Inconclusive assessment, high severity

H Inconclusive - Medium

Status: Unsure
CVSS: Medium
Inconclusive assessment, moderate severity

I Inconclusive - Low

Status: Unsure
CVSS: Low
Inconclusive assessment, low severity

πŸ’‘ Note: Categories provide consistent grouping based on status Γ— severity combinations, while Priority numbers reflect the actual sort order within and across categories. Use categories for thematic grouping and priorities for work sequencing.

πŸ’‘ Note: Categories provide consistent grouping based on status Γ— severity combinations, while Priority numbers reflect the actual sort order within and across categories. Use categories for thematic grouping and priorities for work sequencing.