Enterprise / high-volume plans available on request.
Anonymous users get 10 free calls/day without a key.
Free resource
CFPB Adverse Action Checklist for AI Credit Models
CFPB Circular 2026-03 (May 5, 2026) confirms lenders using machine-learning underwriting remain fully
responsible under ECOA and Regulation B for specific, accurate adverse action reasons — a "black box" model
is not an excuse. This checklist maps what a compliant explanation actually requires. Enter your work email
to unlock it; it prints cleanly to PDF from your browser.
Enter your work email to unlock the full checklist
Unlock above to view all 20 checklist items across 4 sections.
1 · Model Scope & Applicability
Model inventory for credit decisions — every AI/ML model that contributes to a credit approval, denial, pricing, or limit decision, including models used only as one input among several.
Complex-algorithm flag — models identified as "complex" for ECOA purposes (non-linear, ensemble, or otherwise not reducible to a simple point system) get extra documentation scrutiny under Circular 2026-03.
Vendor model coverage — third-party underwriting or scoring models used in your credit decisions are in scope; "the vendor built it" is not a defense against the explainability requirement.
Alternative data sources — documentation of any non-traditional data (cash flow, rent payment history, behavioral data) feeding the model, since these often raise distinct explainability questions.
State law overlay — check for state-level algorithmic accountability or AI-in-credit rules layering on top of federal ECOA/Reg B requirements in states where you lend.
2 · Explainability Capability
No black-box excuse — documented confirmation the firm can produce specific, accurate reasons for any adverse action, regardless of model complexity — per Circular 2026-03, "too complicated" is not a defense.
Reason-code methodology — a documented, defensible method (e.g. feature attribution) for deriving the principal reasons behind a specific adverse decision from the model's actual internal logic.
Reasons tie to factors actually considered — verification that generated reason codes reflect factors the model genuinely used, not a plausible-sounding but disconnected explanation layered on afterward.
Specificity testing — sample adverse action notices reviewed against the "specific reasons" standard, not generic language like "insufficient creditworthiness."
Explainability tooling validated — if using a technical explainability method (e.g. SHAP, LIME, or a vendor tool), documentation that the method itself has been validated as accurately reflecting model behavior.
3 · Notice Content & Process
Principal reasons disclosed — adverse action notices list the actual principal reasons for the decision, ranked by relative contribution where feasible.
Notice timing compliance — process confirms notices are issued within Reg B's required timeframes regardless of how long the model's explanation-generation step takes.
Consumer inquiry process — a defined path for a denied applicant to ask follow-up questions about the stated reasons, with staff able to answer beyond "the algorithm decided."
Human review availability — a process for a human to review and, where appropriate, override an AI-driven adverse decision on request or exception.
Notice template review — legal/compliance sign-off on notice templates generated from model output, not just the underlying model itself.
4 · Governance & Evidence
Fair lending testing — disparate impact testing on the model's outcomes and on its adverse-action reason patterns across protected classes.
Model change re-testing — a policy requiring explainability re-validation whenever the underlying model is retrained or materially updated.
Complaint and dispute log — a record of consumer disputes tied to AI-driven adverse actions, with resolution and any pattern analysis across disputes.
Named accountable owner — a specific individual accountable for ECOA/Reg B compliance of each AI credit model, able to describe the explainability methodology directly.
Examiner-ready documentation — the model inventory, explainability methodology, and sample notices assembled in a form a CFPB or prudential examiner can review directly on request.
Need this filled in, not just outlined?
We build the explainability test suite, verify your reason codes actually match model behavior, and write the report your compliance file can cite.
This checklist is educational, not legal advice — ECOA and Regulation B obligations depend on your specific
model, product, and lending footprint. Confirm your compliance approach with qualified counsel. See our
disclaimer for more.