Artificial intelligence (AI) is playing an increasingly significant role in mortgage finance. These tools are used across most stages of the mortgage value chain, including marketing, underwriting, property valuation, fraud detection, servicing, and loss mitigation.

When used effectively, AI can speed up underwriting, lower origination costs, and make routine servicing tasks faster and less costly. But AI can also produce inaccurate, inconsistent, or unfair outcomes, particularly for families already in financial distress. For example, it can undervalue a home based on a flawed appraisal model or deny a loan without any explanation.

Despite the growing use of AI tools in mortgage finance, the mortgage industry lacked an industry-specific architecture governing AI. That changed in early 2026, when Freddie Mac and Fannie Mae announced and implemented new AI governance frameworks. The Mortgage Industry Standards Maintenance Organization (MISMO), the industry’s standards-setting body, also released its Framework for Responsible AI in Mortgage Ecosystems (FRAME) toolkit in June.

But establishing governance frameworks is only the first step in ensuring AI tools produce accurate, replicable, and equitable outputs. Whether financial institutions effectively translate the frameworks into practice and whether governance ultimately supports better outcomes for borrowers will determine the success of the frameworks. 

Where the frameworks converge: Protecting the mortgage finance system’s safety and soundness

Each framework asks institutions to know where AI is being used, assign accountability for it, and monitor it. In other words, they all agree on a baseline goal: protecting the mortgage finance system’s safety and soundness. The frameworks also demonstrate that there may not be a single model for AI governance and that requirements may need to be flexible enough to support institutions of different sizes, capacities, risk exposures, and use cases.

Of the three frameworks, Freddie Mac’s Guide Bulletin 2025-16 is the most prescriptive. It requires seller-servicers—or institutions that originate or service mortgages approved by Freddie Mac—that use AI or machine learning in their business to establish enterprise-wide governance policies, executive-level accountability, separation of duties, auditing procedures, and processes to address fairness and bias.

Fannie Mae’s Lender Letter LL-2026-04 takes a principles-based approach and provides more discretion to its seller-servicers when implementing the framework. The letter requires governance policies and risk-management practices based on each institution’s business and use of the technology, without specifying the same level of detail as Freddie Mac’s.

By comparison, MISMO’s FRAME is a voluntary, industry-built toolkit aligned with the National Institute of Standards and Technology’s AI Risk Management Framework. The toolkit is intended to help institutions translate responsible-AI principles into operational practices, with a focus on small-to-midsize institutions that may have fewer resources to build governance frameworks.

Where the frameworks diverge: Uneven attention to fair and equitable outcomes

Freddie Mac’s bulletin explicitly names fairness and bias mitigation as a governance obligation. Fannie Mae’s letter asks institutions to build governance needed to ensure AI use is “trustworthy and ethical.” But Fannie Mae does not name fairness or bias explicitly, nor does it specify what human review of an AI-driven decision should look like in practice. Some attorneys are already advising lenders to build their processes according to Freddie Mac’s stricter standard, because it will likely satisfy Fannie Mae’s principles by default. 

MISMO’s FRAME ties its structure back to fair lending laws—including the Equal Credit Opportunity Act—and disparate impact testing, but the specific, testable artifact that would produce evidence of fairness is one MISMO has deferred to a later stage.

Where the gaps remain: Will governance effectively translate into better outcomes for borrowers?

Having a governance process for AI tools does not mean they will lead to more accurate, replicable, and equitable outcomes for borrowers. These frameworks should align their principles to help lenders and servicers advance measurable outcomes, such as the following:

• faster, more accurate matches between borrowers and homes
• expanded access to credit for creditworthy borrowers currently underserved
• more distressed borrowers reached and kept in sustainable loss-mitigation programs before foreclosure
• stronger household wealth building through both easier entry into homeownership and its preservation

None of the current frameworks require evidence on any of these outcomes, which could help lenders and servicers measure AI’s impact in their processes.

Further, all three frameworks include a “human in the loop” as a safeguard, but none asks how to design effective collaborations between technology and meaningful human judgment, accountability, and oversight.

To effectively govern AI use in mortgage finance, institutions and regulators should evaluate the effectiveness of AI tools and human-AI collaborations against the two outcomes that matter most: whether they help more people become homeowners and whether they help more of them stay homeowners for a longer period to build sustainable wealth.

Building an independent evidence base on AI in mortgage finance 

No single governance framework works for every institution. Mortgage market participants will need to determine how these frameworks fit within their institutions’ particular needs and risk profiles.

Institutions will also have to contend with state requirements. Though most states have not taken action, some have passed laws that affect lending and other consequential uses of AI, such as Colorado’s revised Automated Decision-Making Technology Act (effective January 1, 2027) and Texas’s Responsible AI Governance Act (effective January 1, 2026).

Navigating this will take specialized expertise most institutions cannot easily build, and it will only get harder as more states act and as frameworks evolve.

That’s why Urban’s Housing Finance Policy Center has launched a new initiative on AI in mortgage finance. Our goal is to build an independent and trusted evidence base and serve as a bridge across the mortgage finance ecosystem to answer the questions governance alone cannot.

Ultimately, the success of AI governance should be measured not only by whether institutions implement effective processes but by whether those processes improve outcomes for—and reflect the experiences and voices of—the institutions and households they are intended to serve.