AI for Mortgage Loan Officers: A Practical, Compliant Guide

AI can save loan officers real time on marketing, follow-up drafts, meeting notes and internal checklists. It should stay away from credit decisions, borrower advice and anything your compliance team hasn't approved. Since early 2026, lenders that sell to Fannie Mae and Freddie Mac also have written AI governance rules to follow, so "which tool should I use" now comes after "what does my company allow."

This guide is for loan officers, branch managers and small lenders. It covers what is safe to try, what to avoid, and how to set up simple guardrails. We are not your compliance department or your lawyer, so treat this as a builder's view and run your own policy past the people who own that job.

What changed for lenders using AI in 2026?

Both government-sponsored enterprises now have written AI expectations. <a href="https://www.harrisbeachmurtha.com/insights/fannie-mae-and-freddie-mac-set-new-ai-standards-for-mortgage-lenders/">According to Harris Beach Murtha</a>, Freddie Mac issued Guide Bulletin 2025-16 on December 3, 2025, with an effective date of March 3, 2026. Fannie Mae followed with Lender Letter LL-2026-04 in April.

Here is the timeline in short:

The dates and section number come from Pease Bell's breakdown and from Fannie Mae's own page for the letter.

What do the rules actually require?

In short, lenders need to be able to show how they manage AI. Pease Bell summarizes Fannie Mae's letter as <a href="https://www.peasebell.com/insights/fannie-mae-freddie-mac-ai-ml-governance/">a documented governance program, an annual policy review, the same standards for vendors and subcontractors, and disclosure of AI use on request</a>.

The letter is not limited to underwriting. Harris Beach Murtha notes that it <a href="https://www.harrisbeachmurtha.com/insights/fannie-mae-and-freddie-mac-set-new-ai-standards-for-mortgage-lenders/">does not distinguish between internally developed tools and vendor-provided systems</a>, and covers any AI or ML system used in origination or servicing.

Does this apply to me as an individual loan officer?

The obligation sits with the lender, but you are the one using the tools. If you paste a borrower's file into an unapproved chatbot, you may have created a problem your company has to answer for.

Vendors are in scope too. Asurity writes that lenders remain accountable for the quality, safety and compliance of AI tools they use, so vendor due diligence has to cover how outputs are verified. Existing laws did not go away either. As the MISMO president put it, existing laws already apply to AI.

What can loan officers safely use AI for?

Start with work where the AI never sees a borrower's file and never influences a credit decision. The common thread is that a person reads and edits the output before anyone outside the company sees it.

Good starting points:

• Marketing drafts. Social posts, referral partner emails, newsletters and website copy, reviewed for compliance with your company's advertising rules before publishing.
• Your own meeting notes. Summaries of calls with real estate agents or your team, with borrower identifiers removed.
• Follow-up drafts. Templates for checking in with agents and past clients, edited before they go out. We cover the workflow in how to automate lead follow-up.
• Internal checklists and training. A first draft of an onboarding guide for new processors, or a call script for a team meeting.
• Research and outlines. A starting structure for a homebuyer seminar, with facts checked against real sources.

Real estate agents are working through the same questions on their side of the deal. Our guide to AI for real estate agents shows what they are using, which helps when you are writing for referral partners.

What should loan officers never do with AI?

Keep AI out of anything that decides, advises or discloses. Your company's policy is the final word, but these are the lines we would draw.

1. Don't paste borrower data into a tool your company hasn't approved. That includes names, Social Security numbers, income documents, bank statements and addresses.
2. Don't use AI to tell a borrower whether they qualify, what program fits, or anything about pricing. Those belong to licensed people and your lender's approved systems.
3. Don't send AI-written borrower communication without reading it. One wrong sentence about a loan term is your problem, not the tool's.
4. Don't use AI to screen or steer borrowers. Fair lending rules apply whether a person or a program does the sorting.
5. Don't let AI invent facts. Chatbots can produce confident, wrong statements about guidelines and programs. Check anything factual against the source.

How do you vet an AI tool before using it?

Ask the vendor the same questions your compliance team will ask you. If you are at a larger lender, your compliance team does this for you. If you are at a small shop, you do it yourself.

A short vendor checklist

1. Where does the data go? Is it stored, and for how long? Is it used to train the vendor's models? Get this in writing.
2. Who can see it? Look for access controls, logging and clear terms for subcontractors.
3. Can you turn off training on your inputs? Business plans typically offer this, and free consumer plans often don't. Read the terms for the specific plan you would use.
4. How are outputs verified? Asurity's point about vendor diligence is that <a href="https://www.asurity.com/artificial-intelligence/fannie-maes-ai-governance-requirements-are-now-in-effect-what-lenders-should-expect-from-their-third-party-vendors/">lenders must evaluate how outputs are verified</a>. Ask what review steps exist, and who does them.
5. Can the vendor answer an audit request? Freddie Mac's language is described as more prescriptive on audit and indemnification, so expect lenders to push these questions down to vendors.

Our article on whether it is safe to use AI with customer data goes deeper on data terms and what to look for in a business plan.

How do you set up a simple AI policy for your team?

A one-page policy beats no policy. It also gives your compliance team something to improve, rather than a surprise to discover. These steps follow the shape of the GSE frameworks: ownership, written rules, vendor review and an ongoing check.

1. Name an owner. One person is responsible for AI questions. In a small shop, that might be the owner or compliance officer.
2. List every AI tool in use. Include the ones people signed up for themselves. You can't govern what you haven't found.
3. Sort tasks into three buckets. "Allowed," "allowed with review" and "not allowed." Marketing drafts go in the second bucket. Anything touching borrower data or credit decisions goes in the third unless compliance has approved a specific tool.
4. Require human review. Every AI output that leaves the company is read and approved by a person.
5. Review it on a schedule. The GSE frameworks call for at least an annual review. Tools change quickly, so a quarterly check on your tool list is a good habit.
6. Keep a record. Write down what was approved, when, and by whom. If a partner or investor asks, you have an answer.

This is general guidance, not legal advice. Your lender's own policy and your attorney take priority.

What does this look like inside a real mortgage company?

We build software for East Coast Mortgage, which is also owned by Austin. The system we built includes a loan officer portal, CRM, commissions, calendar and Loan Coach, and you can read the details on the East Coast Mortgage case study.

The guardrail we hold ourselves to there is simple. AI helps people organize, draft and prepare, and a licensed person decides and communicates. Nothing in our tooling is meant to replace a loan officer's judgment, and anything lending related still goes through the people who are licensed to handle it.

The practical lesson is to put AI inside the systems your team already uses, with access rules and logging, instead of leaving it in a browser tab nobody tracks. Controlled tools are easier to govern, easier to explain and easier to improve.

When should a lender build instead of buy?

Buy when a vendor tool already does the job, has clear data terms and passes your vendor review. Build when your process is unusual, your data is sensitive, or you need AI to work inside your own systems with your own rules.

For reference, here are our published starting points:

• Free 30-minute consultation to talk through the problem.
• Operations Review, $7,500, 2 to 3 weeks, credited in full toward a build started within 60 days.
• Automations and integrations from $10,000.
• AI rollout with Claude from $15,000.

If your pain is moving data between documents and systems, our guide to automating data entry from PDFs and invoices covers the pattern. For a broader view of where automation pays off, see business process automation for small business.

A note on scope: we build software and workflows. We don't give lending, legal or compliance advice, and we don't promise any particular outcome from an AI rollout. Every AI feature we design keeps a human reviewer in the loop.

Where should borrowers go with lending questions?

Not to an AI tool and not to this article. This guide has no rates, no loan advice and no guarantees. For lending questions, contact East Coast Mortgage at ecomortgage.com (NMLS 2354674).

The short version

• AI is fine for marketing drafts, your own meeting notes, follow-up templates and internal training, as long as a person reviews the output.
• Keep borrower data, credit decisions, pricing and program advice out of any tool your company hasn't approved.
• Freddie Mac's AI requirements took effect March 3, 2026, and Fannie Mae's Lender Letter LL-2026-04 took effect August 6, 2026, so lenders need documented AI governance, including for vendors.
• A one-page policy with an owner, a tool list, three task buckets and a regular review is a solid start.
• When off-the-shelf tools don't fit, build AI into your own systems with access rules, logging and human review.