As mortgage stakeholders strive to integrate artificial intelligence across their marketing, processing, underwriting and hedging tools, uncertainty remains a dominant theme in 2026.

Fundamentally, the ways in which mortgage originators, lenders and servicers are permitted to use AI — and the ways they can’t, per existing mortgage rules — hasn’t been clarified by industry regulators, says the Mortgage Bankers Association (MBA), which published a white paper this week underscoring the issue.

Rapid adoption of AI in the mortgage industry “presents both significant opportunities and complex legal and regulatory challenges,” the report reads. Noting an “absence of comprehensive federal and state guidance on AI in mortgage lending,” the MBA called on the industry to develop “a unified, principles-based risk management framework.”

Fannie Mae and Freddie Mac published guidance introducing AI governance standards in recent months, but the MBA described these updates as “relatively high-level,” leaving key questions unresolved. They nevertheless reflect the government-sponsored enterprises’ recognition, in the MBA’s assessment, that “companies are, and will continue, using AI in mortgage origination and servicing.”

Overarchingly, many rules and regulations that govern mortgage processes — from who can handle applications and extend offers of credit to the traceability of lending decisions spit from automated workflows — are based on laws written without AI in mind.

The Secure and Fair Enforcement for Mortgage Licensing Act (SAFE Act) passed in 2008, for example, which established a nationwide system for licensing and registering mortgage loan originators, defines an MLO as an “individual” who takes loan applications and is compensated for offering and negotiating the terms an approved loan.

“However, the SAFE Act does not answer the broader question of whether mortgage companies can offer completely human-free loan originations,” the MBA maintains. The Truth in Lending Act and its implementing statute Regulation Z also place emphasis on the term “individual” and require “disclosure of the name” and licensing number of “the person who is the primary contact for the consumer, labeled ‘Loan Officer.’”

The trade association points out that while no federal mandate exists requiring that a “human MLO” lead the origination process, such disclosure requirements, as well as legal definitions of who or what can originate a loan, effectively impose such a mandate.

“Consequently,” said the MBA, “even where an AI system performs most of the loan origination tasks, the consumer should be provided the name and [license ID] for a human MLO assigned to their application.”

But failing to disclose at the outset of loan application process that an AI system is either entirely or partially performing the loan origination work could have lenders and originators running afoul of other consumer protection laws shielding borrowers.

Sections of the Federal Trade Commission Act and the 2010 Dodd-Frank Wall Street Reform and Consumer Protection Act expressly prohibit “unfair or deceptive acts or practices” (UDAPs), including among financial service providers like mortgage companies. Many states have their own versions of UDAP laws.

Assigning a specific MLO or licensed processor to an application creates a reasonable expectation that that individual has some involvement in their application or is available for consultation by the borrower, the MBA points out. If the entire origination is performed by AI, or the MLO has no awareness of the file whatsoever, that could be viewed as deceptive.

“Ensuring that the individual MLO assigned to a particular application is familiar with its contents, has confirmed that any key AI outputs are appropriate, and/or is available to serve as a point of contact for the applicant can serve as a critical check against these risks,” said the MBA. A disclosure indicating that any or all origination processes are performed only by AI, without human involvement, could also mitigate risk, the group said.

Beyond the shifting role — or eventual obsolescence — of human originators, the white paper highlights a range of other concerns raised by advanced AI adoption amid uncertain regulatory treatment.

Fair lending and disparate impact risks may emerge for lenders as AI models trained on historical data may perpetuate or amplify biases in their marketing or origination processes. “Explainability” is another challenge, particularly given regulatory requirements for providing clear reasons when adverse credit decisions are made.

Additional risks noted by the MBA include product steering, data privacy and cybersecurity, and reliance on third-party AI vendors with opaque methodologies or less sophisticated internal control and governance standards.