Ben Anderson: AI Does Not Need a Mortgage License. Your Borrower Still Needs a Human.

In June, the Mortgage Bankers Association published a white paper written by the law firm Orrick, Herrington and Sutcliffe, examining AI-powered mortgage lending through the lens of federal law. It answered a question the industry has been arguing about in conference hallways for two years, and the answer was less exciting than either side wanted.

An AI system does not need a mortgage loan originator license. The SAFE Act applies to individuals, and a model is not an individual. That is the headline most people took away from the paper.

The rest of it matters more. Under the Truth in Lending Act and Regulation Z, lenders have to disclose the name and Nationwide Multistate Licensing System identification number of an individual loan officer tied to the loan. The paper’s conclusion is that those disclosure obligations effectively require a human originator to be assigned to every transaction, whether or not any statute says so in plain words. It recommends lenders keep a licensed originator available to the borrower and involved in oversight, even where AI is doing a substantial share of the work.

So the license question was never the real question. The real question is what the human is for.

The Rules Tightened While Everyone Was Watching The Wrong Agency

This spring produced two regulatory moves that point in opposite directions, and lenders who only tracked one of them are going to have a problem.

The Consumer Financial Protection Bureau finished a rewrite of Regulation B on April 22, stepping back from disparate impact as a fair lending theory, effective July 21. For a lender using AI in credit decisioning, that reads like relief. It is narrower relief than it looks. Disparate impact theories remain live under the Fair Housing Act and under a number of state fair lending regimes, and those are not the CFPB’s to narrow.

Meanwhile, the entities that actually buy most of the country’s mortgages went the other way. Fannie Mae issued Lender Letter LL-2026–04 in April, establishing a governance framework for any approved seller or servicer using AI or machine learning, effective August 6. It requires written policies covering the full life of every AI system in use, review at least annually, the same standard applied to vendors, and disclosure to Fannie Mae on request of what the tools do and what safeguards exist. Freddie Mac’s requirements have been in force since March.

States are moving too. Colorado enacted its Automated Decision-Making Technology in Consequential Decisions law on May 14, with an effective date of January 1, 2027. The MBA paper quotes one of its own members on where this is heading if the industry does not build a shared framework first: 50 different states will regulate us 50 different ways.

What This Means For Buying Decisions

The practical consequence is that a lender’s AI vendor is now part of the lender’s compliance surface. Under the Fannie letter, a lender has to be able to explain, on request, what its tools do and how they are governed. That is not a question you want to send to a vendor for the first time after the request has arrived.

Which is why I think the human-in-the-loop requirement is being framed backwards across most of this industry. It gets treated as a limitation, a piece of legacy regulation slowing down an otherwise fully automated future. Vendors design around it and hope the rules loosen.

They should design for it. A borrower who is told they are talking to AI, who has a named licensed professional attached to their file, and who can reach that person when the conversation stops being routine, is a borrower in a compliant transaction. A borrower who believes a human is reviewing their application when nothing of the kind is happening is the specific scenario the MBA paper warns creates exposure under federal and state consumer protection law.

Where We Drew The Line

We built Cindie as an AI workforce for residential lending, and we drew the boundary in the same place the regulation does, because we were mortgage operators before we were a software company.

Cindie handles engagement, lead conversion, application intake, document collection, processing support and post-close retention. It works around the clock, which is when borrowers actually have questions. It does not decide who gets a loan. Licensed mortgage professionals remain responsible for lending decisions and for regulatory compliance, and the borrower always has a person to reach.

That was not a compromise we made to satisfy a compliance department. It is the product. The repetitive communication and administrative work sitting between lean teams and impatient borrowers is enormous, expensive and almost entirely mechanical. Automating it is worth doing on its own terms. Reaching past it into the credit decision buys a lender risk it cannot price and a borrower an experience they did not consent to.

The industry is going to sort itself into two groups over the next 18 months. One will treat AI governance as paperwork to produce when Fannie asks. The other will treat it as a description of how the product should have been built. I would rather be in the second group before the request arrives than after.

Ben Anderson is the founder and chief executive officer of Cindie, an AI workforce built for the mortgage industry. He also founded Low Rate Co and the mortgage coaching platform Ben Anderson 365, and has originated more than $4 billion in home loans. Connect with him on LinkedIn.

For California Business Journal Disclaimers, go to https://calbizjournal.com/terms-conditions/.