I am a long-time subscriber, and I'm grateful you have been writing about AI's impact on mortgage banking. My compliance officer bought your AI Policy Program. Our Director of Sales is considering getting your manual for AI in mortgage banking sales. So, I hope you will allow this long message from someone who has been in this business for forty years.
We use AI in various areas, including operations, compliance, marketing, underwriting, and disclosures. I'm a CEO and loan officer. My company is in 15 states. I use AI in my loan origination and also in my private life. In my opinion, AI seems like a glorified search engine. But I still have to double-check it because it's throwing errors, and there's no quality control to ensure reliability.
A friend of mine had to hire another compliance manager just to make sure that all AI output is strictly within federal and state guidelines. Instead of AI replacing compliance, he is adding to compliance to keep a watch on AI. And he tells me AI isn't saving him a dime. It is actually costing him more. It's tough enough to book loans in this economic climate; we don't need a new, unregulated technology to confuse us or our borrowers.
And I don't get all the buzz about AI replacing me. It won't replace me. I don't have any borrowers who want to talk to an AI bot. We had an AI agent for a while, and borrowers hung up or logged off. They only want humans, such as me. I go to AI the same way I go to any search engine. It seems more like a consolidator of information, mostly plagiarized, and puts it into a readable, friendly format. This thing just doesn't seem like a revolution that is going to put us all out of work.
A day of reckoning is coming, I promise you. Where's the regulation? Who's liable if AI makes a mistake? Who will reimburse me if I get sued because of an AI error? Why is there no quality control? Why are all these FinTech companies springing up all over the place with AI promotions?
Let's get real. I have no confidence that AI technology is a new type of economic revolution. Maybe it is for the billionaires who run it, since it seems to be a revolution in how fast they can become mega-billionaires.
Will there be a day of AI reckoning in mortgage banking?
COMPLIANCE SOLUTIONS
Policies and procedures for artificial intelligence relating to mortgage banking compliance administration, mortgage loan origination compliance, mortgage servicing compliance, and industry best practices — including, but not limited to, applicable federal and state banking and consumer lending laws.
Sales Manual provides 9 chapters covering the full AI-powered origination funnel in a 90-day implementation roadmap, with clear team roles for fair lending, FCRA, and advertising compliance guidance built into every chapter. Sample AI prompts, an ROI tracking framework, and a compliance glossary.
ANSWER
Your concerns are complex and worthy of a detailed response.
I'll answer in three parts and publish them here weekly, beginning with Part I today.
Part I
- The pattern lenders are experiencing
- A regulatory framework closing in fast
- Who actually stands behind the output?
Part II
- Capital Markets Are Starting to Ask the Same Question
- Public Is Rejecting the Infrastructure, Not Just the Valuations
- Data Center Opposition
Part III
- What is the scariest compliance problem for AI in mortgage banking?
- Why This Looks Like an Implosion, Not a Correction
- Future or Futuristic AI for Mortgage Banking
_________________________________
Distributed after publication of all three articles.
ARTICLES ON AI
In my estimation, confidence in AI is cracking faster than the technology is improving. Furthermore, as you correctly surmise, a widening gap exists between AI adoption and AI accountability in residential mortgage banking.
As an avid reader, you know that I have written extensively on AI and mortgage origination or servicing. You may have attended my talks at conferences or attended our webinars. I am going to be very direct: your concerns are legitimate!
_________________
Here are a few recent related articles:
_________________
THE PATTERN LENDERS ARE EXPERIENCING
I perceive a pattern that has become familiar to compliance professionals in mortgage banking over the past year. A lender deploys an AI tool, often for income calculation, disclosure generation, marketing copy, adverse action notices, or borderline underwriting judgment calls, and so forth, because it is fast, inexpensive, and marketed as production-ready.
The tool performs well enough on routine files to build institutional confidence. Then it produces an output that is wrong in a way that matters: miscalculated qualifying income, a disclosure that omits a required element, a marketing claim that reads as a rate guarantee, or an adverse action notice with a reason code that doesn't match the underlying decision. By the time the error surfaces, which happens in an investor review, a regulator exam, or a QC audit, it is no longer a technology problem. It is a compliance problem, and someone has to fix it, document the remediation, and stand behind the correction.
That's where firms like ours come in. And it is worth asking, plainly,
why lenders would rather absorb the compliance risk of a tool that offers no indemnification at all than rely more heavily on the front end, on licensed professionals who do and who carry meaningful errors-and-omissions coverage to back their guidance.The answer isn't irrational. It's structural, pointing to something larger than a single industry's teething problems with a new technology.
A REGULATORY FRAMEWORK CLOSING IN FAST
Mortgage-related AI activities arrived in a comparatively unregulated environment. That's a fact, but it won't stay that way. Within a single window in 2026, three separate regulatory tracks converged on the same conclusion: AI-generated output in a regulated lending function does not relieve the institution of accountability, and increasingly requires the institution to prove it. To emphasize my point, consider:
· Interagency Model Risk Guidance
In April 2026, the OCC, Federal Reserve, and FDIC issued OCC Bulletin 2026-13, replacing the older model risk framework. Notably, the update excludes generative and agentic AI from its scope – not because those tools are unregulated, but because regulators concluded they need to be governed through separate channels, for instance, compliance management, operational risk, third-party oversight, cybersecurity, privacy, and fair lending, all reporting up to a board or, as applicable, management.
· GSE Governance Mandates
Fannie Mae's Lender Letter LL-2026-04 (April 2026) and Freddie Mac's Guide Bulletin 2025-16 (effective March 2026) now impose formal governance, audit, and security requirements on any approved seller or servicer using AI or machine learning. Those obligations extend to vendors and subcontractors, and lenders must hold third-party AI tools to standards "no less protective" than their own. A lender that cannot document how a vendor's AI reached a given output risks a finding that jeopardizes its ability to sell loans into the secondary market.
· CFPB Explainability Requirements
The CFPB's April 2026 final rule amending Regulation B directly addresses AI in credit decisions. Specifically, fair lending law still requires documented human judgment behind every adverse action, regardless of how the underlying analysis was generated.
Layered together, these frameworks touch nearly every AI function in a mortgage shop, including underwriting, income analysis, HMDA reporting, adverse action notices, marketing, and even AML monitoring. Industry compliance webinars this year have repeatedly flagged the same theme: AI is now "literally everywhere" in the loan process, from prequalification to marketing copy, and lending teams are implementing it faster than their compliance frameworks can keep up.
So there is definitely a gap: adoption is outpacing governance maturity, which is where errors are being generated and where firms doing compliance remediation are seeing the volume of correction work rise.
WHO ACTUALLY STANDS BEHIND THE OUTPUT?
The deeper structural issue, and the one your question centers on, is indemnification, or, put otherwise, the near-total absence of it in the AI vendor market.
My firm has surveyed AI vendor contracts. Our survey included 50 AI vendors offering mortgage banking AI tools. The results are stark.
Only about a third of the AI vendors offer indemnification for third-party IP claims, and roughly one in six commit to full regulatory compliance. Both figures are well below norms for conventional SaaS agreements.
The standard AI vendor contract disclaims output accuracy on an "as is" basis, denies responsibility for output generated from a user's own prompts, and caps whatever liability remains at the fees the customer paid and, if so, the cap is often a fraction of the cost of a single regulatory finding or investor repurchase demand.
Insurance is not filling that gap; if anything, it is widening it. Commercial insurers have begun introducing AI-specific exclusions into general liability, cyber, technology E&O, and professional liability policies. Critically, even where an AI vendor nominally agrees to indemnify a customer, the vendor may carry no insurance capable of funding that promise. The lender's own technology E&O coverage, designed for parties that provide technology services, typically does not respond to claims arising from a lender's use of someone else's AI tool.
I spoke to an insurance executive recently who told me that she considers the use of AI Tools to be at "its own risk" for exactly this reason, because, to use her lingo,
"existing coverage categories were not built around probabilistic systems that generate novel, unreviewable output."
Wow! That's a mouthful. Allow me to translate: current insurance policies weren't designed for AI systems that create completely new and unpredictable content; or, more crudely, insurance rules don't fit AI because AI creates new stuff that can't be checked in advance.
Set against that backdrop, a firm carrying $1,000,000 in errors-and-omissions coverage, such as my firm provides, staffed by professionals whose guidance is a work product they stand behind personally and professionally, is not a redundant layer of cost. It is the risk-transfer mechanism that AI vendors, by contract design, have opted out of providing.
Lenders willing to lean more heavily on AI output and less on that kind of professionally accountable expertise are not eliminating risk; they are relocating it, uncompensated, onto their own balance sheet and regulatory record.
------------------
This article, The Day of AI
Reckoning: Part I, published on September 9, 2026, is authored by
Jonathan Foxx, PhD, MBA, the Chairman & Managing Director of Lenders
Compliance Group, founded in 2006, the first and only full-service mortgage
risk management firm in the United States, specializing exclusively in
residential mortgage compliance.
