The mortgage industry is rapidly embracing AI-powered underwriting, automated income analysis, appraisal reviews, and fraud detection tools—but with that innovation comes a critical question: who’s responsible when the technology gets it wrong? In this episode, David Lykken sits down with industry veterans David Kittle and Arthur Prieston to unpack the growing wave of repurchase demands tied to vendor-driven AI decisions and third-party data integrity issues. The conversation explores how liability is shifting from traditional underwriting errors to technology and vendor oversight, why lenders must rethink how they evaluate AI partners, and how repurchase insurance and vendor certification are becoming essential safeguards in today’s evolving risk environment. If your organization is implementing AI anywhere in the loan manufacturing process, this discussion offers timely insight into the legal, operational, and financial implications every mortgage lender should understand.
[David] Listeners, if you’ve been paying attention to the headlines, we’re seeing repurchases are on the rise. We’re seeing results of a successful business. This is that an increase in production. Joining me today is my good friend, David Kittle, and his and my good friend, Arthur Prieston. I met Arthur through David Kittle. What two special guys we have on the podcast, and we’re gonna be talking about a favorite topic. So David, I’ll let you tee this up, and then we’ll get Arthur in on this discussion as quickly as we can.
[Kittle] Thanks, David and thanks for making time to talk to both Arthur and I. I think it’s pertinent to get him back on a podcast on a recorded call with you because of the amount of the repurchase requests that are occurring now with the GSEs, Fannie and Freddie. But several of them, as the increase goes, it’s around AI and the use of it. And I thought talking to some of the lender members in the mortgage collaborative, they are very concerned about litigation around if one of their vendors using AI makes a mistake that causes a repurchase and Arthur has insurance around that, so I think it’s time to have him on there and talk about how we can protect the lenders out there. So Arthur, welcome, and thanks, and maybe you can speak to that. That’d be great.
[Arthur] Yes. And that’s a great little summary. Appreciate it, David
[David] Good. So Arthur, what can lenders do about it? I know you have an insurance product that helps on this. And what’s your perspective? Are you seeing an increase in lend repurchase demands?
[Arthur] Let’s step back a second. Repurchases are normally associated with underwriting errors that the mortgage banker itself has made, and what AI is doing is changing the location of that risk. It’s not eliminating risk, but it’s moving it away from the underwriter normally making the error, not detecting a fraud, and it’s now tied to data integrity and third-party systems. So it’s moved away from the lender in terms of where the potential fraud could exist. It’s not necessarily eliminating it. It is reducing it. But it can and we speak very generally about AI. There are many different types of AI, we all know that, and there are all different types of vendors that sort of are AI-influenced as opp- as opposed to pure AI, and it all still boils down to data integrity, accuracy, and whether or not the system is reliable and every vendor is different. And we’re quick to adopt, both in this industry as well as in our lives, technology, and then we really don’t know whether or not we’re assuming risk or increasing it or even finding a way to detect it. Great point. And that’s really where what’s happening. Before, mortgage bankers have always looked at, protecting themselves with third-party detection systems that are then filtered through the eyes of an underwriter, and that’s fair, and that’s where repurchase responsibility or liability, I should say, lies. Now it’s shifting to choosing which vendor should you choose in order to do your outsourced underwriting, for instance, or appraisal review, for instance, or income calculations or debt analysis or credit analysis and so it’s not so much that that there’s going to be an increase of repurchase because of AI, but there’s going to be an increase in liability of these companies that are providing this information to the lender. The lender’s still ultimately responsible, but they can now look- Yeah … to these vendors.
[David] I remember there was a lawsuit a number of years ago that, I’m not sure what court it was, but it was certified that a vendor that renders and then provides services, even if it’s an LOS, can be implicitly implicated in a problem if it’s at all can be pointed towards that solution, that, that technology. And now we are, to your point, understanding and seeing AI increasingly more. And it’s understandable. We’re all trying to cut costs in this industry. We have been for years. We’re trying to develop more accuracy in our underwriting, which in theory, and I think in practice, AI has a better chance of doing it. But I think the mistake is people are relying on it more and more rather than doing the necessary checks and balances. And that’s where I think we’re the– it’s on both parties. It’s on both parties, which ultimately, like you said, it’s on the lender.
[Arthur] To comment on that is to take a look at the leg-legal relationship between the vendor and the lender. And if the lender is prudent, they make sure that the agreement, the use agreement between the vendor and the lender contains rep and warrants themselves. Because if they’re going to represent to the lender that this particular property is worth this value or this income calculation is correct, they should be rep and warranting that because it has to be relied upon to the detriment of the lender, who then will have exposure to repurchase if their income calculations are wrong. So in the agreements themselves, they should require rep and warrants, and in many instances they’ve added our insurance to that provision, basically saying the vendor is insured by Lloyd’s of London in the event of any error that has resulted in a repurchase request.
[David] Mr. Kittle,I’ll let you jump in on this discussion, and any questions you have for Arthur. Comments.
[Kittle] Not necessarily questions because I’m pretty familiar with what he does, but I think to his point, are you analyzing the agreement before you sign it as a lender with what vendor you’re signing with, especially if they’re using AI now, especially ’cause they’re using it in underwriting or in any capacity that you’re doing. Are you really looking at your agreement? Do you understand what that artificial intelligence is doing for you? And Arthur can certainly help any company with that, and then to put his insurance on top of it gives you great comfort going forward.
[Arthur] I’m gonna comment also on the fact that it’s not– And it’s also very m-much about the insurance in terms of protecting the lender. The lenders, how are lenders choosing between vendors? How do you choose between one, one, one vendor that does A OCR reviews, writing support, et cetera, and another one that does the same thing, okay? So how do you choose between the two of them? And it’s not so much whether or not they have insurance, it’s more about whether you can study the standards and see whether or not they’re in compliance, what would be acceptable standards, error rates, detection rates. Remember, AI is all about patterns, identifying patterns, and then making inferences. That’s what we’re working with today. That’s called weak AI, frankly. That’s the actual term for that. And every vendor is different. They have different levels of pattern recognition and the ability to identify whether or not this is a fabrication, a hallucination, whether or not there’s been some manipulation of data, which can, more sophisticated players are– sophisticated players out there are getting closer and closer to being able to do that. So really, it’s about identifying standards and then applying those standards to your choice of vendor. That’s what we do. You don’t get insurance unless you meet certain standards, and we’ve been developing-
[David] So you help… you’re like a backstop in this process of evaluating the vendors then, ’cause you’re actually the insurance policy that sits there ultimately that can take the risk.
[Arthur] Yeah, I don’t know about a back… The insurance is clearly a backstop. But really it’s about the certification. So if you have a UL, we all know what U-UL stands for when it’s, when you have it on a, a hair dryer, okay? That’s a standard, and the same thing holds true with choosing a vendor. And when you have a Lloyd’s certification, you get that imprimatur, if you will that basically gives some level of comfort to a lender, and not so much comfort from just certification, but also with insurance. So you have certification on the upfront, insurance on the backside, and that’s how you get to choose a-appropriately and prudent, especially for your stakeholders in your company, prudently, to determine that particular vendor I’m about to use as a, to replace my underwriters, for instance, they meet all the qualifications to obtain a policy. And if they can do that, then that’s a plus as well. Okay. So when, how, tell us more, how do people get this insurance? What’s the cost of the insurance? Is it m- is it the cost lower as a result of having AI in there? I mean- Some would think that- I wouldn’t necessarily say the cost is lower, but someone’s gotta pay for the insurance, right? But the cost is lower indirectly because you’re having less and less liability associated with it. And if there is a problem … I’ll give a good example. Recently there was a situation where income calculations were dependent upon a minor receiving his late father’s Social Security benefits. The AI n- didn’t necessarily identify how old that minority was, and as, the minor was, and as a result, they got a put back because the rule should’ve been in place that it is a minor, double-check when that minor becomes emancipated. And so that’s a rule, being eligible for, to be able to collect that income. Can no longer collect it because they’re now- Okay, it can work both sides. It can work both ways. So in that instance, whose fault was that? It wasn’t the AI’s fault. So it’s a matter of who designed it, how the rules were created, why they missed that, did they fix it? And how often do things like that happen? That’s part of our certification process. That’s how we look at these particular vendors to see what where they’ve been built, how they’ve been built, and wha- and what their pattern recognition testing has been over the last five years. And many of them are so new they have no information, so that you have to you have to be somewhat wary of that.
[Kittle] So think about what you just said there, David, that he actually can assist the lender in evaluating the vendor. It’s the same process he goes through doing rep and warranty insurance. You just can’t call up and say, “I want a rep and warranty policy.” Arthur evaluates the company, your underwriting policies, your buyback requests, and evaluates that particular company before he will even take the company on just for rep and warranty insurance. So this can be an add-on to rep and warranty. It can be included with it together, and it’s a great evaluation process so that he can do for the lender.
[Arthur] A lender can also call us at any given time and say, “Okay we’re about to hire a vendor for this purpose. Are they certified by you guys?” And we start from there. A lender could also get a policy itself for all of its vendors. That’s one way of doing it as well. And then of course, vendors can call us up directly. We’re at azpinsurancespecialists.com, and they’re, just go into that, and you’ll be able to contact us and our group, and we’ll be able to set you up in terms of initial discussions about your needs and we are Lloyd’s of London brokers, and we’re exclusive in this area, and we essentially are able to basically design whatever policy you need in a bespoke fashion. So thanks for the time. Really appreciate it, David.
[David] You bet. So good to be here with you. I love the, what you’re doing for the industry, Arthur, and have been for a long time, and this product is really timely. And Mr. Kittle, thanks so much for bringing it to our attention.
[Kittle] Oh, you’re welcome, David, anytime. It’s been a pleasure to be back on here with both you and Arthur.
[David] Yeah. Arthur, how can people reach you?
[Arthur] Again, azpinsurancespecialists.com. Take a look at that site. That’s it. Got some contact information.