The FinReg Frontier: AI and Machine Learning in Consumer Finance

Podcast: The Consumer Finance Podcast

Episode: The FinReg Frontier: AI and Machine Learning in Consumer Finance

Host: Chris Willis, Troutman Pepper Locke

Overview

This episode explains how U.S. regulators evaluate AI and machine learning models used in credit underwriting and fraud detection. It outlines the CFPB’s evolving expectations on fair lending, explains why those expectations continue to matter despite political change, and highlights the growing role of state regulators and AI laws.

How CFPB Expectations Evolved

Beginning in 2022, the CFPB signaled that lenders using automated decisioning models should assess less discriminatory alternatives (LDA). If a model shows disparate impact, institutions are expected to evaluate whether a similarly predictive model could achieve the same business goal with reduced impact on protected classes.

This expectation moved from public speeches into supervisory exams and was formally stated in a 2024 CFPB comment letter to the U.S. Treasury.

2025 Supervisory Highlights: Key Signals

In January 2025, the CFPB released Supervisory Highlights summarizing its approach to AI-driven models. Core themes included:

→ Fair lending considerations embedded across the full model lifecycle

→ Representative training data and careful variable selection

→ Clear business justification and documentation

→ Disparate impact testing and LDA analysis

→ Use of open-source de-biasing tools to compare alternative models

The CFPB also expressed skepticism toward highly complex models with thousands of similar variables, citing explainability and adverse action challenges. Certain categories of alternative data, such as education, occupation, and criminal history, continued to face regulatory resistance.

Supervision Over Enforcement

Regulatory pressure largely took the form of supervisory examinations rather than enforcement actions. Model changes and remediation occurred through exam findings and supervisory feedback.

Why Oversight Extends Beyond the CFPB

Even with uncertainty at the federal level, fair lending scrutiny remains active:

→ State regulators, including the New York Department of Financial Services, conduct rigorous statistical fair lending exams

→ State AI laws, such as the Colorado AI Act, impose duties to prevent algorithmic discrimination in high-risk systems

→ Existing state laws in Massachusetts and California already support similar expectations

→ Private litigation continues to present reputational and legal risk

Long-Term Fair Lending Risk

The Equal Credit Opportunity Act carries a five-year statute of limitations. Model decisions made today remain subject to future regulatory review, reinforcing the need for long-term, consistent governance rather than short-term regulatory shifts.

Alternative Data and Fraud Tensions

The episode highlights a sharp policy shift around alternative data. Earlier encouragement to use alternative data to serve credit-invisible consumers gave way to heightened skepticism. The discussion argues that certain alternative data sources remain accurate, predictive, and valuable when used responsibly.

A second unresolved issue involves fraud prevention. Highly detailed adverse action explanations may weaken fraud controls by exposing detection logic to organized criminal networks. The episode suggests a more balanced approach that aligns fair lending principles with effective fraud mitigation.

Key Takeaways

→ LDA analysis is now a central expectation for AI underwriting and fraud models

→ Fair lending governance remains essential despite political transitions

→ State regulators and AI laws continue to expand oversight

→ Long-term compliance planning matters more than short-term regulatory cycles

→ Model design must balance fairness, explainability, and fraud resilience