As AI takes over mortgage approvals, who’s watching the machines? 

Thinking big about residential real estate success requires a big-picture perspective. Industry Decoded features industry experts who can enrich your understanding of issues affecting the industry as a whole.

The views expressed in this column are solely those of the author.

Eight months ago, I predicted that AI-driven mortgage approval systems would dominate the industry by the end of 2025. At the time, the prediction sounded bold — even a bit futuristic. But today, it feels all but inevitable.

What began as modest experiments — such as using AI to trim processing times or automate fraud checks — have accelerated into something transformative. Increasingly, the decision regarding who qualifies for a mortgage is shaped not by a human underwriter, but by an algorithm parsing thousands of variables at lightning speed.

The efficiency revolution

Notorious for requiring stacks of paperwork, repetitive verifications and endless handoffs, mortgage lending has long been defined by inefficiency.

AI promises to break that slow, costly and error-prone cycle. Instead of a processor manually combing through pay stubs, bank statements and credit reports, an algorithm can analyze them all in seconds. It doesn't get tired, it doesn't miss things and it can flag anomalies across thousands of loans simultaneously.

Recent pilots and product launches show the impact:

• 40–60% reductions in post-close defects, with fewer costly corrections

• 3–5 days shaved off funding timelines, speeding closings

• Approval speeds more than twice the industry average

For borrowers, this means faster answers and fewer headaches. For lenders, it means lower costs and higher throughput in a margin-starved environment.

Expanding the net for borrowers

In addition to efficiency, AI offers inclusivity. Self-employed workers, contractors and gig earners have often fallen through the cracks in traditional underwriting, which favors W-2 employees — but AI can change that. By analyzing cash-flow data, rental payment histories and even utility bills, algorithms can build a more nuanced picture of a borrower's ability to repay.

A prime example: Fannie Mae and Freddie Mac can now use VantageScore 4.0, a scoring model that incorporates rental and telecom payment histories (with Buy Now/Pay Later reporting still to come). This shift could unlock as much as $1 trillion in lending capacity and potentially extend homeownership to 5 million additional households.

If paired with AI's ability to process nontraditional data at scale, this could be one of the most important breakthroughs in expanding access to credit in decades.

The shadow of bias

AI models are trained on historical data — and in U.S. housing, that history carries the weight of discrimination.

Studies by Brookings and MIT have suggested that automated systems can undervalue homes in majority-Black neighborhoods or penalize applicants based on incomplete or "noisy" credit data. A model may not intend to discriminate, but intent doesn't matter if outcomes perpetuate inequity.

The industry argues that AI reduces human bias — and in some respects it does, as algorithms don't possess personal prejudices. But they also don't possess human judgment or context. Machines can apply rules faster, but they can't yet question whether the rules are fair.

That paradox — AI as both equalizer and divider — is the central tension shaping the industry's future.

The watchdog retreats

Normally, this is where regulators would step in. For the past decade, the Consumer Financial Protection Bureau (CFPB) has been the primary watchdog ensuring that lenders comply with fair lending laws.

But in 2025, the CFPB's authority has come under siege. Its director was fired earlier this year, operations were briefly halted and Congress has proposed slashing its budget nearly in half. Its ability to act decisively remains hobbled.

While other regulators monitor developments, none have the same mandate to set clear standards for fairness in AI underwriting.

The limits of self-regulation

Without federal guardrails, lenders are left to police themselves. Some are investing heavily in fairness audits, bias testing and explainability tools. Others are less deliberate, prioritizing speed to market and assuming any issues that pop up can be resolved later.

But once harm is done, it's hard to undo — and when inequities surface, reputational damage spreads quickly. The industry risks fresh waves of criticism if AI — marketed as a force for fairness — is found to perpetuate discrimination.

A fork in the road

Done right, AI could make lending faster, cheaper and more inclusive, helping millions of households who have historically been overlooked. Done wrong, it could hardwire bias into the mortgage system for decades.

The decisions lenders, policymakers and technology providers are making will shape that outcome. Transparency, explainability and fairness need prioritization as much as cost savings and speed.

While the embrace of broader credit models shows what's possible, this is only one piece of a much larger puzzle. AI isn't just coming to mortgage lending — it's already here. AI is already delivering measurable gains in speed, cost savings and borrower experience. But it's also arriving in a cloud of uncertainty, with the regulatory framework unsettled and the risks of bias still real.

The machines are watching us. The real question is: Who's watching them?

Coby Hakalir has been a leader in the mortgage industry for almost three decades. He currently leads the mortgage banking and mortgage tech division for T3 Sixty, one of real estate's most respected management consultancies, and resides in Northern California. (Note: Real Estate News is an editorially independent division of T3 Sixty.)