Since October 1, 2025, mortgage lenders and mortgage-backed securities issuers in the United States that rely on automated valuation models (AVMs) to value a borrower’s home must have policies, procedures and control systems that meet five federal quality control standards. The rule was issued in July 2024 by six agencies (the OCC, the Federal Reserve Board, the FDIC, the NCUA, the CFPB and the FHFA) and published in the Federal Register on August 7, 2024. It is one of the first binding US rules aimed squarely at statistical and machine-learning models that set the value of people’s homes.
What an AVM is, in the rule’s words
The rule defines an automated valuation model as “any computerized model used by mortgage originators and secondary market issuers to determine the value of a consumer’s principal dwelling collateralizing a mortgage.” The definition says nothing about the method. A hedonic regression, a repeat-sales index and a gradient-boosted model trained on millions of sales are all treated the same way.
That is deliberate. AVMs estimate a home’s value from recent sales, property features and location data, and the techniques keep changing. Our overview of AI in real estate explains how they work and why they struggle with unusual homes and thin markets. The agencies chose not to regulate a technique. They regulate the controls around whatever model a lender uses.
The five quality control standards
Covered institutions must make sure the AVMs they use, whether built in-house or bought from a vendor, adhere to standards designed to:
- ensure a high level of confidence in the estimates produced;
- protect against the manipulation of data;
- seek to avoid conflicts of interest;
- require random sample testing and reviews; and
- comply with applicable nondiscrimination laws.
The first four come straight from the 2010 Dodd-Frank Act, which added a section on AVMs to federal appraisal law. The fifth was added by the agencies. The rule’s preamble links it to the Equal Credit Opportunity Act and the Fair Housing Act, which already ban discrimination in credit and in residential real estate transactions, including valuations.
The rule does not set numeric thresholds such as a maximum error rate or a required confidence score. The agencies said they wanted to give institutions flexibility to set controls that fit their size, risk and the models they use, and to keep the rule workable as modeling technology changes.
Who and what is covered
The standards apply when an AVM is used in two kinds of decisions:
- Credit decisions: whether and on what terms to originate, modify, terminate or otherwise change a mortgage, including changing the limit on a line of credit such as a HELOC.
- Covered securitization determinations: whether to waive an appraisal for a loan that may be sold to a secondary market issuer, and structuring, preparing disclosures for or marketing initial offerings of mortgage-backed securities.
The property must be the consumer’s principal dwelling: a one-to-four-unit home, including condominium and cooperative units and manufactured homes. Vacation homes and investment properties are outside the rule.
What the rule does not cover
The exclusions matter as much as the scope:
- Appraisers’ own use of AVMs. A certified or licensed appraiser may use an AVM while preparing an appraisal, but under professional standards the appraiser’s conclusion must stand on its own. The agencies excluded this use.
- Reviews of completed valuations. Using an AVM to check an appraisal or evaluation after it is done is not covered, however long after the original valuation the review happens.
- Portfolio monitoring. Watching the value of collateral in an existing loan book is not a credit decision, so it is outside the rule.
Evaluations, the cheaper alternative to a full appraisal that banks may use for some transactions, are covered when an AVM is used to prepare them.
Why the nondiscrimination standard matters for AI
Home valuations have a documented history of bias, and models trained on past sale prices can carry those patterns forward. Neighborhood-level features can act as proxies for race even when race is never an input, the problem described in our explainer on algorithmic bias. By writing compliance with nondiscrimination laws into AVM quality control, the agencies made fair-lending testing a standing requirement for these models rather than something checked only after a complaint.
In practice, that means lenders need to test whether their models’ errors differ systematically between neighborhoods or groups of borrowers, and need to be able to explain what they did about it.
A worked example: a HELOC limit increase
Imagine a borrower asks a credit union to raise the limit on a home equity line. The credit union uses a vendor’s AVM to estimate the home’s current value instead of ordering an appraisal. Because changing a credit limit is a credit decision on a principal dwelling, the rule applies. The credit union should be able to show that:
- it set a minimum confidence level for estimates it will rely on, and routes low-confidence results to an appraisal or evaluation;
- the data feeding the model cannot be edited by loan staff who benefit from a higher value;
- people with a stake in the loan’s approval do not choose or tune the model;
- a random sample of AVM values is regularly compared with appraisals or sale prices;
- error patterns are checked across neighborhoods and borrower groups.
None of these steps requires the credit union to build its own model. It must, however, understand and control the vendor model it relies on.
What to watch next
- Supervisory findings from bank examiners in the first full exam cycles under the rule.
- Whether vendors publish fairness and accuracy testing to help smaller lenders comply.
- How the rule interacts with appraisal waivers at Fannie Mae and Freddie Mac, which increasingly rely on models.
- State-level action on AI in housing that adds requirements on top of the federal standards.
Key takeaways
- The federal AVM quality control rule has applied since October 1, 2025.
- Six agencies issued it in July 2024; it was published in the Federal Register on August 7, 2024.
- Lenders and securitizers must ensure AVMs meet five standards, including compliance with nondiscrimination laws.
- The rule is technology-neutral and sets no numeric accuracy thresholds.
- Appraisers’ own AVM use, post-valuation reviews and portfolio monitoring are excluded.
Frequently asked questions
What is the AVM quality control rule?
It is a 2024 federal rule from six US financial agencies requiring mortgage originators and secondary market issuers to have controls ensuring the automated valuation models they use meet five quality control standards. It took effect on October 1, 2025.
Does the AVM rule apply to Zillow’s Zestimate?
The rule regulates how mortgage lenders and securitizers use AVMs in credit decisions and securitization, not consumer websites that publish home value estimates. A portal’s estimate is covered only if a covered institution relies on that model for a covered decision.
Does the rule ban AI home valuations?
No. It allows any modeling method, including machine learning, as long as the institution using it has policies and controls that meet the five standards and follow nondiscrimination laws.




