Fair lending and AI credit models: how to defend one

September 15, 2026 · 5 min · Free, with full transcript

AI credit models score applications in seconds and can approve borrowers a human officer would skip, but they can also rebuild race from proxies like zip code, college name or phone carrier. This episode takes the position that the risk is not the automation itself, it is the failure to test and document. A negligent model is worse than a biased human; a disciplined one is better and provable. You will learn the difference between disparate treatment and disparate impact, how a less discriminatory alternative search creates a legal defense, what belongs in a model risk file to meet the business necessity standard, and what CFPB, OCC and DOJ enforcement has cost, including Trustmark's roughly $9 million redlining settlement and over $150 million from the DOJ redlining initiative.

Listen to the podcast

4 min

Chapters

Key takeaways

  • Run a less discriminatory alternative search on every credit model, and document what you found and adopted.
  • Assume removing race does not remove bias, because zip code, college and phone carrier act as proxies.
  • Keep a model risk file listing every variable, its business justification, and proof it predicts default.
  • Ask your data science team to show the less discriminatory alternatives they tested and rejected; silence is a red flag.
  • Cross-check vendor fairness claims, including Google's fairness libraries, against independent research.
Read the full transcript

Host:Welcome to the Leaders Insights Podcast. Today's episode, Building Credit Decisioning Models That Survive Fair Lending Scrutiny. A single fair lending violation can wipe out every efficiency gain your fancy AI model ever produced. So why is every bank racing to automate credit decisions anyway?

Expert:Because the upside is real. A model can score a loan application in seconds instead of days, and it can approve people a human loan officer would have never looked at twice. The problem is that the same model that expands access can quietly build a bias engine, and regulators have gotten very good at spotting it.

Host:Let's back up. When you say fair lending scrutiny, what specifically are the regulators looking for?

Expert:Two flavors. First, disparate treatment. That's when you deliberately treat protected groups differently, like coding race into the model. Nobody's dumb enough to do that anymore. The real danger is the second flavor, disparate impact. A neutral-looking rule that still hits a protected group harder. Your model never sees race, but it uses zip code. And zip code in America is basically race with a mailing address.

Host:So the model doesn't need to know someone's race to discriminate against them.

Expert:Exactly. That's the trap execs miss. They think, we removed the protected variables. We're clean. You're not. The model finds proxy, Zog. A proxy is a stand-in variable that correlates with the thing you removed. Shopping habits, the name of the college, the phone carrier. Strip out race, and the model reconstructs it from crumbs.

Host:That sounds unfixable. If the model will always find a proxy, how do you build one that survives an audit?

Expert:You test for it relentlessly, and you document everything. The core technique is something regulators actually endorse. Yee, you run what's called a less discriminatory alternative search. You take your model, and you ask, is there a version that's slightly less accurate, but meaningfully fairer? If one exists and you didn't use it, you're liable. If you searched, found it, and adopted it, you've got a defense.

Host:Who's actually doing this well?

Expert:Upstart, the consumer lender, publishes fairness testing on its models and has taken it public with regulators. They report expanded approval rates for Black and Hispanic applicants versus traditional scoring. Google AI has pushed tooling for this too. Their fairness libraries claim they can cut disparate impact substantially. But worth noting, Google sells the software, so cross-check their numbers against independent work. MIT Technology Review ran a good piece showing these tools help but don't eliminate the problem. A bad data pipeline poisons everything downstream.

Host:Say the CFPB, that's the Consumer Financial Protection Bureau, shows up at your door. What actually saves you?

Expert:Documentation. Boring, unglamorous documentation. You need a model risk file that shows every variable, why it's there, what business reason justifies it, and proof you tested alternatives. The legal standard is business necessity. You have to show the variable predicts default, and there's no less discriminatory way to get that signal. If your answer is the model liked it, you're already writing the settlement check.

Host:How big are those checks, realistically?

Expert:Trinity Bank, sorry, Trustmark, settled a redlining case for around $9 million a few years back, and that's on the small end. The DOJ's redlining initiative has pulled over $150 million in settlements. Their efficiency gain from faster underwriting is maybe a rounding error against that.

Host:Here's the uncomfortable one. Isn't the honest move to just keep a human in the loop and skip the AI risk entirely?

Expert:No. And this is the part people get backwards. Human loan officers are wildly biased. Decades of data show that. A well-tested model is auditable in a way a human's gut never is. You can't subpoena a loan officer's subconscious. The model at least leaves a paper trail you can inspect and correct.

Host:So the AI is riskier to bill but easier to defend.

Expert:If you build it right, the negligent version is worse than a human. The disciplined version is better and provable.

Host:Give me the one thing an exec should do Monday morning.

Expert:Ask your data science team a single question. Show me the less discriminatory alternatives you tested and rejected. If they go quiet, you don't have a model. You have a lawsuit with a training accuracy score.

Host:Sources for today's episode, TechCrunch AI, The Decoder, Google AI, Vendor, AI Lab, KD Nuggets, MIT Technology Review, ours Technica AI. That's it. The AI decision tools are live at MBA-training.com.

Read and go deeper