CFPB Deleted Disparate Impact from Reg B. Here's What Actually Changed for AI Lenders Today.

Today is the effective date for one of the most significant fair lending rule changes in years — and based on the questions compliance teams have been asking, many AI lenders still don’t have a clear picture of what it actually means.

The CFPB’s amendment to Regulation B removes the effects test — commonly called disparate impact — from the rule implementing the Equal Credit Opportunity Act. Effective today, July 21, 2026, statistical disparity alone is no longer enough to establish an ECOA violation under federal law. The Bureau took the position that ECOA is a disparate-treatment-only statute, and deleted every reference to the effects test from the regulation and official commentary.

Here’s what compliance teams at AI lenders need to understand first: the risk didn’t go away. It changed shape.

TL;DR

• The CFPB’s Reg B amendment eliminating the ECOA disparate impact (effects test) takes effect today, July 21, 2026
• Statistical disparity alone no longer proves an ECOA violation under federal law
• What still applies: Fair Housing Act disparate impact (fully intact, enforced by DOJ/HUD), state fair lending laws in CA/MA/NJ and others, GSE contractual requirements, ECOA disparate treatment (use of demographic proxies, intentional discrimination)
• Your adverse action notice requirements are completely unchanged
• AI lenders are not deregulated — they’re now facing a fragmented patchwork of state laws and FHA exposure without a unifying federal ECOA floor

What the Rule Actually Changed

The CFPB published the final rule in the Federal Register on April 22, 2026. The operational change is straightforward: Section 1002.6(a) no longer contains the effects test provision, and the official CFPB commentary no longer references disparate impact as a basis for Reg B liability.

As Cooley’s analysis summarized, the CFPB “deleted every reference to the so-called effects test” and “reframed ECOA as a disparate-treatment-only statute.” Facially neutral criteria are now actionable under ECOA only where they are “intentionally designed or applied as proxies for prohibited characteristics.”

The rule also narrows the “discouragement” prohibition and restricts Special Purpose Credit Programs (SPCPs) — but those changes are separate from the disparate impact piece and deserve their own analysis.

What changed: Under the prior rule, if your credit model produced statistically disparate outcomes for a protected class, regulators could bring an ECOA claim even without demonstrating discriminatory intent. That route is now eliminated at the federal ECOA level.

What didn’t change: Everything else. And everything else is substantial.

The Fair Housing Act Is Untouched

The most important point for residential mortgage lenders: the Fair Housing Act (FHA) disparate impact standard is entirely unaffected by a CFPB rulemaking.

The FHA’s disparate impact liability was affirmed by the Supreme Court in Texas Department of Housing and Community Affairs v. Inclusive Communities Project, Inc., 576 U.S. 519 (2015). The CFPB has no authority to amend the Fair Housing Act. DOJ and HUD enforce the FHA independently, and neither agency changed its fair lending enforcement posture as part of the Reg B rulemaking.

The FHA burden-shifting framework works like this:

1. Plaintiff shows the practice caused or perpetuates disparate impact on a protected class
2. Defendant must show the practice serves a legitimate, nondiscriminatory interest
3. Plaintiff may show a less discriminatory alternative serves that interest equally well

This framework applies to AI credit models, automated underwriting systems, pricing algorithms, marketing targeting, and any other facially neutral practice used in connection with a residential mortgage. The deletion of the ECOA effects test changed none of this.

For non-mortgage lenders: the FHA applies to “dwelling-secured” credit — primarily residential mortgages. If your AI credit product is an unsecured personal loan, auto loan, or credit card, FHA exposure doesn’t apply. But that doesn’t mean you have no disparate impact exposure — which brings us to state law.

State Fair Lending Laws: Now the Default Floor

This is where the practical compliance challenge gets more complex. As Norton Rose Fulbright observed, the “state-law analog framework — independent disparate impact statutes in California, Massachusetts, New Jersey, and a growing roster of other jurisdictions — now operates without a corresponding federal floor or a unifying federal interpretive standard.”

Without the ECOA floor, there is no longer a uniform national standard for disparate impact analysis in consumer credit. Lenders face a patchwork of state-level obligations that vary by jurisdiction, protected class coverage, and enforcement intensity.

California, Massachusetts, and New Jersey have state fair lending statutes that reference disparate impact independently of federal ECOA. California’s Unruh Civil Rights Act and the Fair Employment and Housing Act create multiple avenues for disparate impact claims in credit. The DFPI — which brought the Yotta FDIC insurance enforcement action earlier this month — has signaled aggressive use of the California Consumer Financial Protection Law to fill gaps left by the reduced federal posture.

The practical implication: if you previously ran one disparate impact analysis to satisfy ECOA/Reg B and relied on federal interpretive standards to calibrate your methodology, you may now need to run jurisdiction-specific analyses for each state where you originate loans with its own independent fair lending framework.

ECOA Disparate Treatment: Still Fully Intact

The removal of the effects test does not touch disparate treatment, which remains fully prohibited under ECOA and Regulation B.

Disparate treatment is intentional discrimination — using facially neutral criteria as proxies for protected characteristics. For AI models, this is the more subtle and arguably higher risk in practice. If a model uses zip code as a variable, and zip code is highly correlated with race in your lending geography, and that correlation is documented in your model development process, you may have disparate treatment risk — regardless of what your model’s stated intent was.

The Reg B commentary still makes clear that facially neutral criteria can constitute disparate treatment when they are “intentionally designed or applied as proxies for prohibited characteristics.” For AI models trained on large feature sets, proxy variable analysis isn’t optional — it’s a core component of fair lending compliance. That work didn’t become any less important because disparate impact was removed.

AI bias testing methodologies for fair lending need to address both disparate impact (now mapped to FHA and state law frameworks where applicable) and proxy variable identification for disparate treatment (still required under ECOA). The legal basis has shifted; the operational work largely hasn’t.

Adverse Action Notices: Zero Change

Compliance teams at AI lenders should note clearly: the adverse action notice requirements in Reg B Section 1002.9 are entirely unchanged.

The CFPB’s removal of the effects test had nothing to do with adverse action notice obligations. Lenders using AI or complex credit models must still provide:

• Specific, substantive reasons for adverse action that are the principal reasons the application was denied
• Reasons that applicants can understand and potentially act on
• Notices that satisfy both ECOA Reg B and FCRA Section 615

The CFPB’s 2025 Supervisory Highlights flagged fintech lenders using AI models with hundreds or thousands of variables providing vague, generic adverse action reasons that failed the specificity requirement. That enforcement risk is entirely unchanged by today’s Reg B amendment.

For a complete guide to adverse action notice requirements for AI credit models, see our coverage of FCRA adverse action notice requirements for fintechs using AI.

GSE Contracts: Read Them Before Assuming Relief

For lenders who sell loans into GSE pools (Fannie Mae, Freddie Mac), your seller/servicer contracts impose their own fair lending requirements — and a CFPB rulemaking removing the ECOA effects test does not amend those contracts.

GSE seller/servicer guides have historically referenced disparate impact analysis as a component of seller/servicer compliance obligations. Lenders in the GSE seller/servicer channel should review their specific Fannie Mae Seller/Servicer Guide and Freddie Mac Single-Family Seller/Servicer Guide for fair lending provisions that may impose disparate impact testing requirements independent of what Reg B now says.

What Your AI Fair Lending Program Needs Post–July 21

The post-Reg B world doesn’t simplify AI fair lending compliance. It fragments it. Here’s how to map your obligations:

For mortgage lenders using AI underwriting: Your FHA exposure is unchanged. Run disparate impact analysis against the FHA burden-shifting framework. Document your business justification for any variable showing disparate impact. Test for less discriminatory alternatives. The enforcement agency changed (DOJ and HUD, not CFPB) — the analysis itself didn’t.

For non-mortgage consumer lenders using AI: Your federal ECOA disparate impact exposure is removed. But map your state-specific obligations by origination state, continue proxy variable analysis for disparate treatment, and review GSE contracts if you sell into those pools.

For all AI credit lenders: Adverse action notice requirements are unchanged. Pre-deployment bias testing for disparate treatment should continue. Model documentation should record which variables were evaluated for proxy characteristics. If your fair lending compliance documentation was framed primarily as “disparate impact testing under ECOA Reg B,” update it to reflect the current legal basis — FHA (if mortgage), state law (by origination state), and disparate treatment analysis (all credit types).

So What Does This Actually Mean for Your Program?

The compliance teams that handle today’s effective date well will do three things:

First, update legal basis documentation. Your AI fair lending testing framework probably cites ECOA/Reg B as the primary authority. That citation is now incorrect for disparate impact purposes. Update it to reflect FHA authority (for mortgage), applicable state law (by origination state), and retained ECOA authority for disparate treatment and adverse action notices.

Second, map your state origination footprint. State AI laws tracker 2026 is a useful reference point — many of the same states that have enacted AI-specific regulations are also the states with independent disparate impact liability (California, Massachusetts, New Jersey). If you originate in multiple states, your fair lending compliance program now needs jurisdiction-specific analysis rather than a single ECOA/Reg B framework.

Third, don’t stop testing. The lawyers who read the headline that CFPB removed disparate impact and concluded their fair lending testing program can be scaled back are going to have a bad time. Disparate treatment requires the same statistical analysis as disparate impact — the difference is the legal theory applied to the results, not the work. And with state law and FHA still providing liability exposure, the case for maintaining or expanding testing is stronger, not weaker.

As Venable noted, lenders should be “aware that state fair lending laws and the Fair Housing Act may still impose disparate-impact liability” — and that the CFPB’s change creates a compliance environment with more jurisdictional variation, not less testing need.

The AI Risk Assessment Template & Guide (buy here) includes pre-deployment bias evaluation tools and adverse action notice guidance mapped to the current 2026 AI regulatory landscape, including the post-Reg B fair lending analysis framework.

The Broader Context: State Enforcement Is the Action Now

The CFPB’s rollback of the ECOA disparate impact standard is part of a consistent pattern: as the federal agency retracts from consumer financial protection enforcement, state regulators are stepping in with their own posture.

California’s DFPI, NYDFS, and the Massachusetts AG have demonstrated appetite for consumer financial protection enforcement that doesn’t depend on federal CFPB action. The same dynamic that’s driving state enforcement of UDAAP claims is now operating in fair lending — with state attorneys general and financial regulators pursuing disparate impact theories under their own statutes, with no CFPB coordination or interpretive guidance to anchor the analysis.

For AI lenders operating nationally, July 21, 2026 is an effective date, not a clearing event. The work changes form; it doesn’t go away.