Wirewalk Instituteof AI
gov-330 Expert Waitlist

Model Risk Management for AI

For model risk managers, validators and model owners in banks, insurers and other regulated firms who must extend model risk management to machine learning, LLM applications, vendor models and agents.

The curriculum is being prepared and reviewed. Enrollment is not open yet.

About this course

For fifteen years SR 11-7 anchored model risk management in US banking and far beyond it. In April 2026 the Federal Reserve, OCC and FDIC replaced it with revised, non-binding interagency guidance (SR 26-2, OCC Bulletin 2026-13, FDIC FIL-15-2026) that states generative and agentic AI models are not within its scope, while the UK PRA's SS1/23 and Canada's OSFI Guideline E-23 (effective May 2027) set principles-based expectations that reach AI explicitly. At the same time, firms are deploying machine learning, foundation-model applications and agents that strain classic validation techniques. This course teaches you to apply proportionate model risk management to all of them, including how to govern the generative and agentic systems the US guidance carves out, where firms must rely on their broader risk management and third-party, fair lending and consumer protection obligations rather than on the model risk guidance itself.

You will compare the three supervisory frameworks and decide what counts as a model, including vendor models, LLM-based tools and agentic workflows. You will design inventory and materiality tiering, assign roles across the three lines, assess conceptual soundness for ML, and set documentation and reproducibility standards. Validation modules cover benchmarking, sensitivity, stability and explainability for ML; task-specific evaluation, hallucination and prompt-sensitivity testing for LLM applications; and approaches for black-box vendor models. You will set up ongoing monitoring with population stability and outcomes analysis, manage overrides and change control for retraining and model version upgrades, address fair lending, and report aggregate model risk to senior management and the board. Read the primary supervisory texts for binding detail; this course is educational, not legal or regulatory advice. The Wirewalk Institute certificate attests to completion and assessment of this course only.

What you will be able to do

  • Explain what the 2026 US interagency guidance changed relative to SR 11-7, including its generative and agentic AI carve-out, and compare it with PRA SS1/23 and OSFI E-23
  • Determine whether an ML system, LLM application, vendor model or agent falls within model risk scope and tier it by materiality
  • Assess conceptual soundness and documentation for machine learning models
  • Design an independent validation plan for an LLM application, including task-specific evaluations and prompt-sensitivity tests
  • Validate vendor and black-box models using outcomes analysis, benchmarking and contractual evidence
  • Set up ongoing monitoring, override tracking and change control for retraining and model version updates
  • Report aggregate model risk and model limitations to senior management and the board

Syllabus

6 modules · 18 lessons · about 20 hours

Module 1: The Supervisory Frameworks in 2026

What regulators now expect and how far it reaches into AI.
3 lessons
  1. From SR 11-7 to SR 26-2 and OCC Bulletin 2026-13: What Changed and What Stayed
  2. PRA SS1/23 and OSFI Guideline E-23: Principles-Based Expectations That Name AI
  3. Is It a Model? Scoping ML, Vendor Tools and the Generative and Agentic AI Carve-Out

Module 2: Inventory, Tiering and Governance

The structure that decides how much scrutiny each model gets.
3 lessons
  1. Model Inventory Design: Required Attributes and Interdependencies
  2. Materiality and Complexity Tiering That Drives Validation Depth
  3. Roles Across the Three Lines and Model Risk Appetite

Module 3: Development Standards for AI Models

What developers must justify and document before validation.
3 lessons
  1. Conceptual Soundness for Machine Learning: Data, Features, Algorithms and Hyperparameters
  2. Reproducibility: Experiment Tracking, Model Registries and Versioned Data
  3. Foundation-Model Applications: What the Deploying Firm Can and Cannot Document

Module 4: Independent Validation

Effective challenge for ML, LLM applications and vendor models.
3 lessons
  1. Validating ML Models: Benchmarking, Sensitivity, Stability and Explainability
  2. Validating LLM Applications Outside the US Guidance: Task Evaluations, Hallucination Rates and Prompt Sensitivity
  3. Validating Vendor and Black-Box Models

Module 5: Ongoing Monitoring and Change Control

Keeping validated models valid as data, vendors and versions shift.
3 lessons
  1. Performance Monitoring, Population Stability and Outcomes Analysis
  2. Overrides, Challenger Models and Documented Limitations
  3. Change Control for Retraining, Prompt Changes and Upstream Model Version Updates

Module 6: Special Topics and Board Reporting

Fair lending, agentic systems and communicating aggregate model risk.
3 lessons
  1. Fair Lending Model Risk: Adverse Action Reasons and Less Discriminatory Alternatives
  2. Agentic Systems: Tool Use, Autonomy and Compounding Error
  3. Reporting Aggregate Model Risk to Senior Management and the Board

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 25 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 50 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

More in Governance, Risk and Compliance

All 14 courses in this track
gov-001 Free intro Available

AI Rules in Plain English: What Everyone Should Know

Start with existing US law, learn when state and sector requirements need review, and distinguish voluntary frameworks from binding obligations. Practice checking an AI-assisted task and raising a useful concern.

2 h · 7 lessons Free intro
gov-101 Basic Waitlist

AI Compliance Essentials for Every Employee

A plain-language course on using AI at work without creating legal, privacy or reputational problems. Covers the rules that already apply to your job, the difference between approved and shadow tools, what data you may paste, how to check and disclose AI output, and how to recognize and report an AI incident.

5 h · 16 lessons Join the waitlist
gov-201 Intermediate Available

AI Governance and Risk Management: NIST AI RMF and ISO/IEC 42001

A practitioner's course in building AI governance that works: inventory, intake, risk tiering, impact assessment, lifecycle controls and leadership reporting, anchored to NIST AI RMF 1.0, its Generative AI Profile and ISO/IEC 42001. Ends with a 90-day program build plan and ready-to-adapt templates.

14 h · 18 lessons See price at checkout