AI Regulation Tracker  /  Regulation in force

Federal AVM Rule in Force: Five Standards Every Algorithmic Home Valuation Must Meet

Six federal agencies set quality-control standards for the algorithms that value homes, and the rule took effect October 1, 2025. It reaches the AI valuation models behind lending decisions, and it adds a fifth factor most people miss: compliance with fair-lending law. Here is what is binding and the first move for anyone whose deals run on an automated value.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published June 28, 2026  ·  Last updated June 28, 2026

On October 1, 2025, a joint final rule from six federal agencies (OCC, Federal Reserve, FDIC, NCUA, CFPB, and FHFA) took effect, setting quality-control standards for the automated valuation models (AVMs) that mortgage originators and secondary-market issuers use to value a consumer’s principal dwelling. Issued under section 1473(q) of the Dodd-Frank Act and published at 89 FR 64538, the rule requires covered institutions to govern their AVMs against five standards, including a new fifth factor on compliance with applicable nondiscrimination laws. Primary source: Federal Register, "Quality Control Standards for Automated Valuation Models" (Document 2024-16197).

The algorithms that value homes now have a federal rulebook

For years, an automated valuation model could decide what a home was worth with very little regulation aimed at the model itself. That changed on October 1, 2025, when a joint final rule from six federal agencies took effect and put quality-control standards around AVMs for the first time.

An AVM is a computerized model that a lender or secondary-market issuer uses to determine the value of a consumer's principal dwelling. In many current implementations, that includes machine-learning approaches trained on large property datasets. The rule does not ban these tools. It does something quieter and more lasting. It says that if you use one to make a lending or securitization decision, you have to govern it, and you have to be able to show how.

For real-estate professionals, lenders, appraisers, and the proptech firms building these models, the takeaway is the same. The valuation algorithm is now a regulated input, not a convenience that sits outside the rules.

Who issued it, and where it comes from

The rule was adopted jointly by the Office of the Comptroller of the Currency, the Federal Reserve Board, the Federal Deposit Insurance Corporation, the National Credit Union Administration, the Consumer Financial Protection Bureau, and the Federal Housing Finance Agency. That six-agency coordination is the signal. This is not one regulator's pet project. It is a coordinated federal standard.

The authority traces to section 1473(q) of the Dodd-Frank Act, which amended title XI of the Financial Institutions Reform, Recovery, and Enforcement Act of 1989 to require quality-control standards for AVMs used in valuing the real estate collateral behind mortgage loans. The final rule, published at 89 FR 64538 on August 7, 2024, implements that mandate. Congress called for these standards back in 2010. The rule that finally carries them into force is the one that matters now.

The five standards every covered AVM has to meet

Under the rule, an institution that uses AVMs in covered transactions must adopt policies, practices, procedures, and control systems that meet five quality-control standards. The model has to:

• produce estimates with a high level of confidence
• protect against the manipulation of data
• seek to avoid conflicts of interest
• require random sample testing and reviews
• comply with applicable nondiscrimination laws

The first four come straight from the underlying statute. The fifth is the one the agencies added, and it is the one to read twice. Existing fair-lending law already applied to how an institution used an AVM. By naming nondiscrimination as a standalone factor, the agencies created an independent requirement to build controls that specifically target discrimination risk in the model itself, not just in the human decision around it.

That is the heart of the AI angle. An algorithm trained on historical housing data can absorb and repeat the patterns in that data, including biased ones, and produce values that disadvantage protected groups without anyone intending it. The fifth factor tells covered institutions they cannot treat that risk as someone else's problem. It has to be a named, documented control.

Who is actually covered, and why the definitions are the whole game

The rule is short. The substance lives in the defined terms, and they are broad.

It reaches "mortgage originators" and "secondary-market issuers." A mortgage originator is, roughly, anyone who for compensation takes a mortgage application, helps a consumer obtain or apply for a mortgage, or offers or negotiates mortgage terms, plus anyone who advertises that they do. A secondary-market issuer is any party that creates, structures, or organizes a mortgage-backed securities transaction.

The trigger is using an AVM to make either a "credit decision" (whether and on what terms to originate, modify, or change a mortgage) or a "covered securitization determination" (such as waiving an appraisal requirement, or structuring and marketing an initial mortgage-backed securities offering). The definitions also reach business-purpose loans when the mortgage secures a consumer's principal dwelling, even though those loans sit outside the Truth in Lending Act. Using an AVM only to monitor or verify the performance of existing loans is outside the rule.

The practical instruction is to not assume you are out of scope. The defined terms catch more participants than a quick read suggests.

A flexible rule is not a soft one

The agencies deliberately declined to dictate exactly how to build these policies and controls. A community lender and a national securitizer should be able to calibrate to their size, risk, and complexity. That flexibility is real, and it is easy to misread as optional.

It is not optional. The rule does not create its own enforcement mechanism, which sounds like relief until you read why. The consequences for falling short are the consequences that already attach to the laws the agencies oversee: examinations, supervisory findings, fines, penalties, and enforcement actions. A fair-lending failure that runs through an ungoverned AVM is still a fair-lending failure. The rule did not invent the liability. It removed the excuse for not having controls against it.

What to do first

Start with scope, because the definitions decide everything. Determine whether your firm, or a third party or affiliate acting for it, fits "mortgage originator" or "secondary-market issuer," and whether you use AVMs in a credit decision or a covered securitization determination. If yes, inventory every AVM in the stack, including vendor models, and map each to the five standards.

Then give the fifth factor its own line. Document the specific controls that test the model for disparate outcomes, rather than a general statement that you comply with fair-lending law. The agencies pointed institutions to existing model risk management guidance as a starting point, so you are not building from nothing. The exposure is not using an AVM. It is using one you cannot show you controlled, on the day an examiner or a claimant asks how the number was produced.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.