By: Dr. Joshua Scarpino, Vice President Information Security

August 3, 2026

AI in mortgage origination is often framed as a tradeoff between efficiency and trust, automation and control, scale and compliance. That framing misses the real issue. The real risk is not automation itself; the risk is unobservable decisioning. Systems that act without making their reasoning, options, or pathways visible to either the borrower or the loan officer create exposure that compounds quickly.

The U.S. Department of the Treasury said as much in its 2024 report on AI in financial services, flagging opacity as a direct threat to both regulatory compliance and consumer trust. In lending, that’s not an abstract concern. It shows up in regulatory standing, borrower confidence and conversion. If you can’t explain a system, you can’t defend it.

“Black box” isn’t just a model problem

People tend to reduce black-box risk to model explainability – a data science issue. It’s bigger than that. It shows up in three places:

• Why a borrower is contacted at a specific moment
• What options were considered or ruled out
• Whether a human can actually step in.

Hide any of these, and the system goes dark – not just to regulators, but to the people who are supposed to use and trust it. The CFPB has been explicit on this point, creditors still owe borrowers specific, accurate reasons for adverse actions, even when a complex model produced the decision Model complexity is not a defense under the Equal Credit Opportunity Act.² Most AI deployments don’t fail because the model is bad. They fail because everything around the model is invisible.

Build for visibility, not just accuracy

A more effective approach treats AI not as an autonomous decision-maker, but as part of an observable engagement system. That’s the thinking behind TrustEngine. Rather than collapsing complexity into a single output, the system surfaces recommendations, structures borrower interaction, and keeps loan officers continuously informed throughout with auditability. Transparency is not layered on after the fact; it is built into how the system is designed. The result is not less automation, but more visible automation that can be understood, monitored, and acted upon.

Borrower choice is a governance mechanism

One of the most important design choices: don’t route borrowers down an AI-picked path. Give the loan officer structured options instead – refi opportunities, home equity strategies, purchase-timing scenarios– built from the underlying signals.

That sounds like a UX decision. It’s actually governance. Presenting alternatives makes the system’s logic self-explaining with no technical translation required. It also flips engagement from something that happens to the borrower into something they opt into, which produces cleaner behavioral signals that are easy to measure and audit. The borrower stops being a target and becomes a participant.

The loan officer is the control, not the checkbox

The role of the loan officer is critical. Human-in-the-loop is often treated as a compliance safeguard, but in a well-designed system, it operates as a control mechanism. NIST’s AI Risk Management Framework (AI RMF 1.0) reinforces this position. It identifies accountability, transparency, and explainability as core characteristics of trustworthy AI and emphasizes that effective oversight depends on humans who can interpret system behavior and intervene meaningfully across the AI lifecycle.³ This isn’t new to financial services, the Fed and the OCC’s model risk management guidance introduced “effective challenges” over a decade ago – critical, informed pushback capable of catching model limitations and forcing changes.

• Loan officers inside an AI-driven system are doing exactly that. They’re not rubber-stamping outputs, they’re interpreting, adapting and intervening. TrustEngine supports this by giving loan officers clear visibility into why a recommendation was triggered, what options exist and traceability on how the borrower engaged. This visibility enables three functions: Interpretability– turns the interaction into language the borrower actually understands
• Intervention – lets LOs adjust outreach based on context the model can’t see
• Auditability – lets the organization trace outcomes back through the full chain of signals and decisions.

A human in the loop only reduces risk if that human can see what the system is doing, otherwise it’s just a name on a form.

Transparency doesn’t cost you performance

When borrower choice and loan officer visibility are combined, the system shifts from a black box to what can be described as a glass box. Decision pathways are visible, engagement logic is traceable and outcomes become explainable in plain language. Transparency is embedded in the system’s behavior rather than reconstructed after the fact, and the shift does not come at the expense of performance.

There is a persistent industry assumption that adding transparency, choice, and human oversight slows things down. The data says the opposite. When borrowers understand why they are being contacted and can choose how to proceed, engagement becomes more relevant and less intrusive. Conversion improves because interactions are aligned with borrower intent rather than imposed on it. Organizations also benefit from clearer compliance narratives, reduced model risk exposure, and stronger adoption among loan officers who see the system as a support tool rather than a replacement. Trust, in this context, is not a constraint on performance; it is a multiplier.

References

1. U.S. Department of the Treasury, Artificial Intelligence in Financial Services (December 2024). https://home.treasury.gov/system/files/136/Artificial-Intelligence-in-Financial-Services.pdf
2. Consumer Financial Protection Bureau, Consumer Financial Protection Circular 2023-03: Adverse Action Notification Requirements and the Proper Use of the CFPB’s Sample Forms Provided in Regulation B (September 19, 2023).https://wayback.archive-it.org/23481/20250327014847/https://www.consumerfinance.gov/compliance/circulars/circular-2023-03-adverse-action-notification-requirements-and-the-proper-use-of-the-cfpbs-sample-forms-provided-in-regulation-b/
3. National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (January 2023). https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf
4. Board of Governors of the Federal Reserve System and Office of the Comptroller of the Currency, Supervisory Guidance on Model Risk Management, SR Letter 11-7 / OCC Bulletin 2011-12 (April 4, 2011). https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm