DayBlink Consulting Partner and Cyber Security Practice Lead Michael Morgenstern contributed to OWASP’s “AIVSS Scoring System for Agentic AI Core Security Risks” (v0.8).
As agentic AI systems move from experimental use into sustained enterprise deployment, they introduce a new class of security risks not fully addressed by existing application security or AI governance frameworks. This publication establishes a standardized vulnerability scoring system — the AI Vulnerability Scoring System (AIVSS) — specifically designed to quantify and prioritize the unique risks posed by agentic AI. Grounded in the NIST AI Risk Management Framework and informed by the Cloud Security Alliance’s MAESTRO architecture, the document provides both a taxonomy of the ten core agentic AI security risks and a mathematical methodology for assessing them consistently across organizations.
Key Takeaways:
- A catalog of ten core agentic AI security risks — including tool misuse, access control violations, identity impersonation, and goal manipulation — with detailed attack scenarios for each
- A risk amplification model that builds on CVSS v4.0 baselines, applying ten agentic-specific factors (such as autonomy, memory persistence, and multi-agent coordination) to reflect how agentic capabilities multiply the severity of traditional vulnerabilities
- Alignment with NIST AI RMF’s Govern, Map, Measure, and Manage functions, enabling organizations to integrate AIVSS into existing risk management workflows
- An implementation guide covering lifecycle integration, release gates, and continuous improvement to support ongoing security assessment of agentic AI deployments
Read the full publication here: Link