AI in underwriting is not a new topic. Automated desktop underwriting has been an industry resource for decades. But advances in machine learning — and the emergence of platforms designed to move underwriting and other mortgage manufacturing functions into increasingly automated "black boxes" — are making the rules of engagement less clear.
In the name of consumer protection, federal and state regulators — including those in Michigan, New York, California, Colorado, and elsewhere — have introduced legislation and issued guidance addressing the growing use of AI tools. A careful reading of what is out there reveals less in the way of new requirements and more of a reminder that lenders must evaluate AI tools through the lens of fair lending, discrimination, bias, transparency, disclosure, and consumer protection.
On May 5, 2026, the CFPB issued Circular 2026-03, advising that lenders using complex algorithms, such as machine-learning underwriting models, remain fully responsible under ECOA and Regulation B for providing specific, accurate reasons for adverse action. This places a clear responsibility on lenders to understand how these algorithms work so that any adverse lending decision generated by an AI-driven platform can be translated into specific, accurate reasons for denial. Lenders cannot simply claim that the black box "told us to do it." The Circular also makes clear that proprietary or "uninterpretable" models do not excuse compliance.
Lenders should:
- Understand the AI systems they use, including their algorithms and outputs.
- Ensure adverse action notices clearly and specifically explain the reasons for a denial or other adverse determination.
- Document how model outputs are translated into consumer-facing explanations so fair lending reviewers can independently validate the reasoning.
If you're already using AI-driven underwriting, now is the time to test, validate, and document how those systems reach their decisions. If you're evaluating a new platform, ask tough questions about explainability, compliance, and fair lending — not just speed and efficiency. Faster, better, and cheaper may be part of the sales pitch, but they are not the whole compliance picture.
