Model Risk Management in the Age of AI: Balancing Innovation, Ethics, and Compliance

Sam Mylinarski from the Federal Reserve Bank of New York explores how model risk management (MRM) is evolving to meet the challenges posed by modern AI and generative models. Hear how companies can adjust risk frameworks to address bias, explainability, hallucination, misuse, and prompt sensitivity, while aligning with national guidance like the NIST AI Risk Management Framework.

Key Takeaways:

1. MRM Frameworks Must Evolve for GenAI
   Traditional validation approaches must expand to include LLM-specific risks like hallucinations, prompt testing, and grounding.
2. Explainability is a Regulatory Imperative
   Regulators expect transparent, interpretable models—even with complex AI. This drives tooling and documentation standards.
3. Model Governance Requires Integrated Platforms
   Integrated data science platforms help manage reproducibility, validation workflows, and evidence generation in a unified system.