Model risk management in 2026: A banker's guide to the revised interagency guidance

Summary

The post presents the April 17, 2026 revision to model risk management guidance as a shift toward risk-based, principles-driven oversight, replacing several earlier issuances from the Federal Reserve, FDIC, and OCC. It identifies five operational changes: materiality tiering, end-to-end lifecycle governance, reproducible effective challenge, continuous drift and performance monitoring, and application of MRM principles to GenAI and agentic systems. Its proposed response is a Databricks reference architecture built on Unity Catalog, Delta Lake, Lakeflow Declarative Pipelines, Feature Store, MLflow, Model Registry, Model Serving, and assurance tools. The design makes tiering metadata-driven and turns lineage, validation, approvals, monitoring, documentation, and retirement records into evidence generated during normal model work. The stated conclusion is that a unified substrate can reduce integration and manual evidence work, allowing regulatory changes to become configuration exercises rather than multi-quarter programs.

Context

The revised guidance introduces risk-based tiering, lifecycle-wide governance, reproducible effective challenge, continuous monitoring, and principles-based coverage for GenAI and agentic systems. The post describes traditional implementation as a multi-quarter effort involving inventory migration, validation rewrites, monitoring pipelines, documentation, vendor-model onboarding, and separate AI workstreams, with fragmented tooling consuming scarce MRM capacity.

Approach / What changed

The post proposes a Databricks reference architecture on a governed lakehouse substrate. Unity Catalog anchors inventory, ownership, tiering, access, and lineage, while Delta Lake, Lakeflow Declarative Pipelines, Feature Store, MLflow, Model Registry, Model Serving, Lakehouse Monitoring, AI Gateway, and workflow tools support evidence generation across the model lifecycle. Tier changes and control adjustments are represented through metadata, policies, and configuration rather than migrations.

Takeaways

• Unity Catalog is positioned as the governance anchor, allowing inventory, ownership, materiality tier, access, and end-to-end lineage to resolve through one governed object graph.
• The architecture maps lifecycle work to evidence: data quality metrics, feature histories, experiment runs, validator sign-offs, promotion records, drift alerts, model documentation, and retirement trails.
• A first-pass automation layer can run standardized MLflow evaluations for GenAI systems, such as toxicity, groundedness, and PII leakage, before work reaches a model risk center of excellence.

Related reading