OWASP AIVSS

CVSS was designed for conventional software and does not capture what makes agent vulnerabilities dangerous, such as how much autonomy an agent has or how widely its actions propagate. AIVSS addresses that gap.

AIVSS considers factors such as:

• Base severity: Technical impact similar to traditional scoring.
• Autonomy: How independently the affected agent acts.
• Tool access: The power of tools the agent can use.
• Propagation: How far effects can spread across agents and systems.
• Context: Environmental factors that raise or lower real-world risk.

The project released version 0.8 of its scoring system for OWASP agentic AI core security risks, in co-publication with AIUC-1 and other OWASP projects.

Consistent scoring helps teams prioritize agent risks and communicate severity to leadership and auditors.

How PointGuard AI Helps

PointGuard AI publishes its own AI Security Severity Index (AISSI) to score real-world incidents on the AI Security Incident Tracker, and AI Security Posture Management prioritizes AI risks using context such as agent autonomy and data access.

Learn More

• OWASP AIVSS Project
• OWASP AIVSS v0.8 Scoring System for Agentic AI Core Security Risks
• OWASP Top 10 for Agentic Applications