Responsible AI by Design for GenAI: Milliman’s Model Governance Framework

Milliman’s work on a Model Governance Framework for actuaries is designed to translate regulatory and ethical principle into operational validation methods for Large Language Models, Retrieval Augmented Generation systems and Agentic AI used in actuarial and insurance analytics. Bridging the EU AI Act, NIST AI RMF and U.S. state-level requirements (e.g., Colorado SB 21-169), the framework delivers step-by-step model validation and audit routines grounded in measurable technical metrics and documentation standards. The framework is structured around the following elements: audit and accountability, transparency and compliance, bias mitigation and fairness techniques, robustness and security protocols and ongoing monitoring. This presentation explains how governance and data science teams jointly built a reproducible assurance pipeline that transforms responsible AI by design into a verifiable technical practice. 

What you will learn:

• Discover how Milliman translated regulatory compliance requirements into measurable validation routines for GenAI. 
• Gain an understanding of practical methods within the governance framework by examining their application across various use cases in the insurance industry. 
• Learn how to apply this sector-specific assurance pipeline beyond insurance into broader AI governance contexts.