Compliance & Model Risk in an AI World
AI stretched the definition of a model, and financial institutions now run vendor engines, generative tools, and chatbots that influence customers and decisions. This course applies model risk management to AI at principle level, with regulatory statements dated as of September 2026. You'll learn to inventory and tier AI uses, validate models you didn't build, test fairness and adverse-action reasons, and keep documentation that survives an exam. You'll finish with a one-page inventory line and a ten-question checklist for every model review.
Course details
Duration
1h 13m 04s
Lectures
15
Video
1.25 hours
Quizzes
1
Level
Beginner
Course Description
Model risk management has been part of financial-services supervision for over a decade, built around scorecards, pricing, and capital models. AI stretched it: vendor engines, language models, prompts, and chatbots now shape customer outcomes. This course shows how to inventory, validate, monitor, and govern them at principle level.
What you'll learn
- What counts as a model now, and which AI uses belong in the inventory
- The model lifecycle under AI, and validation against your own evaluation set
- Explanations that reflect real model behavior, and vendor models managed as your own
- Fair-lending testing, the less discriminatory alternative search, and adverse-action reasons
- Chatbot controls, data governance, and the model file, decision record, and change log
- Three-lines roles, risk tiering, drift monitoring, and model incident remediation
Who it's for
Model owners, model risk and validation teams, compliance and fair-lending staff, internal auditors, and business leaders at financial institutions who own, challenge, or audit AI uses.
15 short video lessons in three modules: Model Risk Fundamentals; Fairness, Consumers, and Decisions; and Operating the Framework.
This course is educational and is not legal, compliance, or regulatory advice. Your model risk policy and counsel govern.
Module 1: Model Risk Fundamentals
-
1.1 What Counts as a Model NowFrom scorecards to language models — the inventory question.
In this lesson:
- Define a model broadly enough to capture AI uses
- Decide which AI uses belong in the model inventory
- Start the inventory conversation in your institution
-
1.2 The Model Lifecycle
Development, validation, implementation, use, monitoring, retirement — and where AI stresses each stage.
In this lesson:
- Describe the six stages of the model lifecycle
- Identify where AI models stress each stage
- Apply lifecycle controls to a generative or vendor model
-
1.3 Validation for AI Models
Conceptual soundness, outcomes testing, and the limits of validating a model you did not build.
In this lesson:
- Apply the three validation questions to AI models
- Design outcomes testing for probabilistic and generative outputs
- State the limitations of validating vendor models honestly
-
1.4 Explainability in Practice
What can and cannot be explained, and what a regulator will accept.
In this lesson:
- Distinguish the kinds of explanation a model can provide
- Match the explanation to the audience and the requirement
- Validate that explanations reflect the model's actual behavior
-
1.5 Vendor and Third-Party Models
Due diligence, contractual rights, and ongoing monitoring obligations.
In this lesson:
- Apply model risk expectations to vendor AI
- Secure the contractual rights model risk management needs
- Run ongoing monitoring of vendor models as your own
Module 2: Fairness, Consumers, and Decisions
-
2.1 Fair Lending in an AI Context
Protected classes, proxies, and disparate impact testing at a working level.
In this lesson:
- Describe the fair-lending exposures AI models create
- Run disparate impact testing at a working level
- Search for and document less discriminatory alternatives
-
2.2 Adverse Action and Reasons
Producing accurate, specific reasons from complex models.
In this lesson:
- State the standard adverse-action reasons must meet
- Build and maintain the reason-generation process for a complex model
- Test reasons as a model output
-
2.3 Consumer Communications and Chatbots
Accuracy, disclosure, escalation to a person, and complaint handling.
In this lesson:
- Apply consumer-protection principles to AI-generated communications and assistants
- Set the controls a customer-facing assistant needs
- Handle complaints involving AI-generated content
-
2.4 Data Governance for Models
Lineage, quality, permissible use, and retention.
In this lesson:
- Apply data governance principles to the data AI models consume and produce
- Assess permissible use of data for a model
- Manage retention for model inputs, outputs, and records
-
2.5 Documentation That Survives an Exam
The model file, the decision record, the change log, as of September 2026.
In this lesson:
- Assemble the model file examiners expect for an AI model
- Keep decision records that reconstruct individual outcomes
- Maintain a change log that ties every change to its re-validation
Module 3: Operating the Framework
-
3.1 Roles: First, Second, and Third Line
Who owns, who challenges, who assures — with AI-specific responsibilities.
In this lesson:
- Assign AI model responsibilities across the three lines
- Recognize the AI-specific duties each line picks up
- Spot the role gaps that AI programs commonly leave
-
3.2 Risk Tiering AI Uses
Low, medium, high by consequence — the review each tier needs.
In this lesson:
- Tier AI uses by consequence using consistent criteria
- Match validation, review, and monitoring intensity to the tier
- Re-tier when use, scope, or evidence changes
-
3.3 Monitoring and Drift
Performance decay, model changes by vendors, and the triggers for re-validation.
In this lesson:
- Design a monitoring plan for an AI model
- Detect drift in inputs, outputs, fairness, and explanations
- Set the triggers that move a model from monitoring to re-validation
-
3.4 Incidents and Remediation
When an AI-influenced decision is wrong at scale — containment, customer remediation, reporting.
In this lesson:
- Run a model incident from detection through containment
- Remediate affected customers systematically
- Meet reporting expectations and close the loop into the framework
-
3.5 Your Model Risk Checklist
The one-page inventory and tiering template, the checklist, and your next model review; the finale.
In this lesson:
- Use the one-page inventory and tiering template
- Run the model risk checklist on any AI use
- Complete Compliance & Model Risk in an AI World
Compliance & Model Risk in an AI World — Final Exam
Learn at your own pace
Study the materials at your convenience, you can always return to them to repeat
Certificate of completion
Complete every lesson and pass the final exam to earn a RiseCert certificate of completion you can download and share.
Your Name
Compliance & Model Risk in an AI World
RISECERT
CERTIFICATE
For demonstrated knowledge and successful completion of all course requirements for:
MM.DD.YYYY
Certificate Date
Ready to get started?
Enroll in Compliance & Model Risk in an AI World and earn your RiseCert certificate of completion.
Got Questions? We've Got Answers!
Answers to the most common questions about this course.
What is the Compliance & Model Risk in an AI World course?
Compliance & Model Risk in an AI World is a RiseCert online course that applies model risk management to AI in financial services. It covers what counts as a model, validation, explainability, vendor models, fair lending, adverse-action reasons, chatbots, data governance, documentation, three-lines roles, tiering, monitoring, and incidents. It includes 15 video lessons (about 76 minutes) and a final exam.
Who should take an AI model risk management course?
Model owners, model risk and validation staff, compliance and fair-lending professionals, internal auditors, risk managers, and business leaders at banks and other financial institutions who own, challenge, or audit AI uses. It also suits product and operations teams deploying vendor AI, chatbots, or document tools that influence customers or decisions.
What is model risk management?
Model risk management is the discipline of controlling the risk that a model is wrong or misused. It follows a lifecycle of development, validation, implementation, use, monitoring, and change and retirement, with controls at each stage. The course notes it has been part of financial-services supervision for over a decade and shows where AI stresses each stage.
Is a generative AI tool or chatbot considered a model?
Usually, yes. The course's working test asks whether a tool's output influences a customer, a decision, a risk measure, or a report. AI drafting customer letters at scale, chatbots answering customers, and extraction feeding underwriting belong in the inventory, and prompts and configurations count as part of the model. Personal drafting tools are logged as approved tools instead.
How do you validate a vendor AI model you didn't build?
Compliance & Model Risk in an AI World teaches validating it against your own evaluation set and population, treating vendor material as input rather than proof. Undisclosed training data is documented as a limitation with compensating outcome tests, validation holds for a specific version, and contracts need change notice, version holds, explanation access, and exit rights.
What is a less discriminatory alternative search?
It's the documented effort to find whether a model could meet the same business need with less disparate impact, for example by removing proxy features, adjusting weights, or changing thresholds. The course teaches building and testing alternatives, quantifying the trade-off, and recording the decision. As of September 2026, it describes the expectation to search and document as established at principle level.
What is model drift, and how is it monitored?
Drift is when a model stops behaving as it did when validated. The course covers four kinds: input drift, output drift, performance decay, and fairness drift, which can widen even while overall accuracy holds. Monitoring uses metrics with thresholds and owners, distribution tracking, and evaluation-set re-runs, with unexplained shifts treated as unannounced changes.
What will I be able to do after completing the course?
You'll be able to decide which AI uses belong in the model inventory, tier them by consequence, design outcomes testing for text and scores, validate explanations, secure vendor contract rights, run disparate impact testing, build an accurate adverse-action reason process, keep exam-ready documentation, set re-validation triggers, run a model incident, and use a ten-question model risk checklist.
Do I need a data science or statistics background?
No. The course works at principle level in plain language, so you don't need to be a data scientist or statistician, and there are no formal prerequisites. Some familiarity with banking compliance, credit decisions, or risk management helps, because the course builds on those ideas and applies them to vendor engines, chatbots, and generative models.
Does the course cover fair lending and adverse-action reasons?
Yes. Module 2 covers fair-lending exposures from AI, including proxies, historical bias, feature interactions, and differential accuracy; disparate impact testing at a working level; the less discriminatory alternative search; and adverse-action reasons that are specific, accurate, actionable, and demonstrable, produced from validated attributions and tested monthly as a model output.
How is this different from a general AI compliance course?
General AI compliance courses survey laws and frameworks across industries. Compliance & Model Risk in an AI World applies the financial-services model risk discipline to AI specifically: model inventories, validation, vendor contracts, fair lending, adverse-action reasons, three-lines roles, tiering, drift monitoring, and incident remediation, plus the model file, decision record, and change log examiners read.
Is the course content up to date?
Yes. The course dates its regulatory statements as of September 2026 and works at principle level, so it stays useful as specific rules change. It deliberately leaves jurisdiction-specific requirements, statistical thresholds, and reporting timelines to your institution's compliance team, model risk policy, and regulators' guidance, which always govern.
How long does the Compliance & Model Risk in an AI World course take?
The course has about 76 minutes of video across 15 lessons, plus a final exam. Lessons run about five minutes each and are organized into three modules: Model Risk Fundamentals; Fairness, Consumers, and Decisions; and Operating the Framework. Most learners can finish it in one or two sittings.
Is the course online and self-paced?
Yes. The course is 100% online and self-paced, and your access does not expire. You can start, pause, and resume at any time on a computer, tablet, or phone with a modern web browser. There's nothing to install.
Is there a final exam?
Yes. The course ends with a 30-question multiple-choice final exam covering all three modules. You need a score of 80% or higher (24 of 30) to pass, and you can retake the exam as many times as you need. Finish all 15 lessons and pass the exam to complete the course and earn your certificate.
Do I get a certificate of completion?
Yes. After you complete all 15 lessons and pass the final exam, you can download a RiseCert certificate of completion from your account. It is a course completion certificate, not a license, professional certification, or continuing education (CE or CPE) certificate.
Is the course approved for CE or CPE credit?
No. Compliance & Model Risk in an AI World is not currently approved for continuing education credit, including CE, CPE, or CEU credit, by any state, licensing body, or accrediting organization. If you need credit toward a license or professional designation, take a course approved by the body that issues it.
Is this course legal, compliance, or regulatory advice?
No. The course is educational and is not legal, compliance, regulatory, or financial advice, and it is not endorsed by any regulator. It teaches model risk principles, but your institution's model risk policy, its compliance team and counsel, and your regulators' guidance govern decisions about specific models, disclosures, and reporting.
How much does the course cost, and can I enroll my team?
The course costs $99 per learner as a one-time purchase, with no subscription or recurring fees. To enroll a model risk, compliance, or audit team, contact RiseCert through the Contact Us page at risecert.org.
What is the refund policy?
All sales are final and no refunds are issued, as stated in the RiseCert Terms of Use. Please review the course description, lesson list, and these FAQs before purchasing, and contact RiseCert with any questions first.