← All home signals
Governance

The Mortgage Model Audit

Fannie Mae and Freddie Mac now make AI oversight a condition of buying a lender’s mortgages, including, at Freddie Mac, bias audits and an indemnity for AI losses.

What is it?

Fannie Mae and Freddie Mac now set conditions on how the lenders they buy loans from use AI in origination and servicing. Governance becomes a term of sale: approved policies reviewed annually, audits, bias monitoring and, at Freddie, an indemnity for any AI-related loss.

Why now?

Freddie Mac’s AI section took effect on 3 March 2026, and Fannie Mae issued its own framework on 8 April, effective 6 August 2026. Both GSEs now govern lender AI, and Fannie’s version reaches third-party vendors at the same standard as the lender.

Who’s building it?

Mortgage lenders and servicers, their AI vendors, Fannie Mae, Freddie Mac, and law firms reading the new terms.

Freddie MacFannie MaeMortgage lendersLoan servicersAI vendors
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — vendors of AI and machine-learning tools for mortgage lending.
Buyers — lenders and servicers that sell loans to the GSEs and now owe governance.
Gatekeepers — Fannie Mae and Freddie Mac, writing AI terms into contracts.
What’s at stake
Whether governance paperwork tells a buyer anything about whether a lending model is fair or accurate.
Who pays when a lender’s AI causes a loss.
Whether AI vendors can meet the lenders’ governance standards.
What’s unsettled
How the GSEs will check governance they only see on request.
What bias monitoring means when no test is stated.
Whether an indemnity for any AI-related loss makes small lenders avoid the tools entirely.
02 · How the demand matureda progression, not a timeline
Novelty
Lenders bring AI and machine learning into origination and servicing work.
Retention
Freddie Mac makes continued AI use conditional on annually reviewed policies.
Willingness to pay
Lenders accept an indemnity for any AI-related loss as the price of selling loans.
Category
Mortgage AI becomes a governed input, written into the terms of sale.
03 · Where the market goesthe forks, not a prediction
→Governance as the floor: both GSEs’ terms become the standard every mortgage AI vendor builds to, audits and bias monitoring included.
→The indemnity makes smaller lenders drop AI tools rather than carry the loss.
→Fannie Mae tightens toward Freddie’s more prescriptive terms, and the two frameworks converge into one standard.
→A bias finding in a sold loan tests the indemnity in court.
The forks turn on whether the GSEs treat governance as a filing to keep on request, or as something they will actually test in lenders’ models.
04 · The Model Governance deskadjacent market, moving now
Compliance Moves to Code
Governance tooling built into AI stacks.
High-risk AI must produce continuous compliance evidence from August 2026.
Raconteur
The Explained Denial
Automated housing refusals now owe a plain reason.
Colorado’s new law drops its testing and impact-assessment duties.
Colorado Public Radio
The Liability Question
Who carries the loss when AI goes wrong.
EU withdrew its proposed AI Liability Directive in early 2025.
IAPP
05 · The Wire
01
Freddie Mac requires AI audits, bias checks and an indemnity.
Freddie Mac Single-Family Seller/Servicer Guide ↗
02
Fannie Mae followed with its own AI governance framework.
Fannie Mae Single-Family News ↗
03
LL-2026-04 takes effect 6 August 2026 and extends the same governance standards to third-party vendors.
Cooley ↗
04
Freddie Mac’s Bulletin 2025-16 took effect on 3 March 2026.
Harris Beach Murtha (JD Supra) ↗
The panel on this signal
They agreed on
–Both GSEs now make AI governance a condition of doing business, but neither approves lenders’ models itself.
–Freddie Mac’s terms are the more prescriptive of the two.
–The indemnity moves the cost of an AI-related loss onto the lender.
The mortgage quality-control manager disagreed on
–Annual policy sign-off and on-request disclosure check that governance exists, not that a valuation or underwriting model gives the right answer.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.