ACSC AI Model Risk Scoring Formula Toolkit

Run your numbers.

The ACSC AI Co-Chairs’ risk scoring formula helps you evaluate model risk on a zero-to-one scale.

Includes:

• jailbreak rates

• hallucination rates

• poisoning resilience

• attack frequency

• availability indicators
  

All public data. Additional material from the calculations shared on Dr. Maybury's GitHub repository. https://github.com/mtmaybury/AI-Risk-Assessment

AIThreatsVulnerabilities&ResponsesCalculationToolkit.pdf

Download PDF • 966KB