SR 11-7 AI Model Governance Guide

What's Inside

• SR 11-7 Requirements Overview — The three pillars of model risk management: model development and implementation, model validation, and ongoing monitoring
• LLM-Specific Challenges — How non-deterministic outputs, prompt sensitivity, and emergent behaviors create unique governance requirements for LLM-based agents
• Model Development Documentation — What regulators expect in model documentation for AI agents: design rationale, training data provenance, feature selection, and limitation disclosures
• Validation Framework for AI Agents — Effective challenge requirements, independent review processes, and benchmarking methodologies adapted for LLM-based models
• Champion-Challenger Testing — How to run controlled experiments comparing AI agent outputs against baseline models, with statistical significance thresholds
• Ongoing Monitoring Requirements — Drift detection (PSI, KS tests), performance degradation tracking, and trigger-based revalidation schedules
• Model Inventory Management — Maintaining a complete model inventory that includes AI agents, with risk tiering and review cadences
• Audit-Ready Documentation — Templates for model risk reports, validation summaries, and ongoing monitoring dashboards that satisfy examiner expectations

Related Reading

• SR 11-7 and AI: How to Validate LLM-Based Models
• Why Datadog and Arize Are Not Enough for Financial AI
• ECOA Compliance Checklist for AI Lending Agents