A 25-page methodology defining the first empirical vulnerability scoring standard for AI systems. Five dimensions. 174 analyst-verified incidents. Open methodology.
The first vulnerability scoring standard purpose-built for AI systems, derived empirically from 174 analyst-verified incidents — not extrapolated from traditional CVE or CVSS frameworks.
Technical Sophistication, AI Amplification Factor, Enterprise Impact, Detection Difficulty, and Cascading Risk Potential — each calibrated to the unique threat surface of AI systems.
Comprehensive taxonomy covering AI-native vulnerability classes including agent compromise, memory poisoning, model extraction, prompt injection, and cascading system failures.
A single normalized composite score enabling direct comparison across vulnerability categories, prioritization queues, and portfolio-level risk aggregation for security teams and insurers.
Empirical scoring reveals that AI-native vulnerability classes — agent compromise, cascading failures, and memory poisoning — score significantly higher than categories existing frameworks were designed to handle.
Full scoring methodology, factor definitions, and weighting rationale published openly. Practitioners can reproduce scores, adapt the framework to their environment, and audit scoring decisions.
The highest-scoring vulnerability categories in our dataset are the AI-native ones that existing frameworks were not built to differentiate: Agent Compromise (avg 80.5), Cascading Agent Failures (avg 80.3), and Memory & Context Poisoning (avg 75.0).
ASI-VSS v1.0 — AI Security Intelligence LLCOperational guide for security teams, CISOs, and risk analysts. Translates the full scoring methodology into actionable workflows for vulnerability triage and prioritization.
Download PDFUnderwriting translation for carriers, actuaries, and insurance executives. Maps VSS scores to premium adjustments, coverage conditions, and portfolio aggregation.
Download PDFScore any AI vulnerability against the five-dimension framework. Interactive tool for security practitioners and risk analysts to apply the VSS methodology in real time.
Open CalculatorThe complete 25-page white paper including scoring methodology, dimension definitions, category benchmarks, and empirical findings from 174 analyst-verified AI security incidents.
Download White Paper (PDF)For enterprise licensing inquiries: contact@aisecurityintelligence.com