Adverse Action When AI Is Involved: Demand a Meaningful Explanation, Request Model Evidence, and Appeal

Why this matters now

As lenders increasingly use machine learning and other algorithmic models to underwrite loans, deny credit, or change existing account terms, consumers who receive an adverse action notice can — and should — demand specific, usable explanations and the documentary evidence that supports the decision. The Consumer Financial Protection Bureau has reminded lenders that using AI does not excuse them from providing specific reasons for adverse actions; generic checkbox notices that don't reflect the actual factors used are not sufficient.

This article explains the legal baseline, practical steps to request a meaningful explanation and model evidence, sample language you can send, likely pushback (and how to respond), and next steps if the lender refuses to explain or correct a decision.

What the law requires — the quick legal baseline

Equal Credit Opportunity Act (ECOA) and its implementing Regulation B require creditors to give applicants a written notice of adverse action that either lists the specific principal reason(s) for the action or tells the applicant they have the right to request a written statement of specific reasons. The regulation and its official commentary make clear that the consumer must receive reasons that are specific and that "relate to and accurately describe the factors actually considered or scored by a creditor."

Separately, when a consumer report (a credit or specialty report) contributed to the decision, the Fair Credit Reporting Act (FCRA) requires the user of that report to provide the consumer with the CRA’s name and contact details, a notice of the consumer’s dispute rights, and the ability to get a free copy of the report within a limited time window. These FCRA disclosures complement ECOA/Regulation B; together they give you both the report information and the right to ask for precise reasons.

The CFPB has specifically warned creditors that adverse-action rules apply equally when underwriting relies on complex algorithms and AI — and that lenders must give explanations that actually reflect the model’s principal reasons for the decision rather than a generic nearest-match checkbox. In plain terms: using a black-box model is not a legal escape hatch.

Practical steps to demand a meaningful explanation and request model evidence

1. Read the notice carefully and preserve everything. Save the adverse-action notice, any pre-adverse notices, emails, screenshots of the application, and the exact date you were notified.
2. Get the consumer report and score details. If the lender says a consumer report or credit score was used, request the free copy of the report (you typically have a 60-day window to request a free copy after an adverse action) and ask for the credit score and the key factors that contributed to it. FCRA requires these disclosures.
3. Demand a specific principal reason in plain language. If the lender checked a generic reason (for example, "other" or "insufficient projected income") ask them to explain how that maps to identifiable, addressable behavior or data (for example: "transactions at cash advance merchants on these dates," or "inferred monthly income of $X because of direct-deposit patterns"). The CFPB says reasons must "relate to and accurately describe" the factors actually considered.
4. Request model evidence and documentation. Ask the creditor (politely but firmly) to produce, as applicable:
   • the principal reason(s) identified by the model and how they map to observable data points in your application or files;
   • a description of the model inputs used for your decision (types of data, not necessarily full training datasets);
   • a plain-English explanation of how the principal reason caused the adverse action and what, if anything, you could change to improve your outcome;
   • the identity of any third-party data providers or scoring vendors used.
   Many lenders have internal model explainability tools or mapping tables that show how model outputs translate to decision reasons; CFPB guidance expects lenders to use those tools when they exist.
5. Ask for a human review and reconsideration. Request an officer-level reconsideration or manual review of your application and provide any additional evidence (pay stubs, bank statements, explanations of unusual transactions) that could change the model’s inputs.
6. Use deadlines and be specific. Ask for the requested information in writing within a reasonable deadline (for example, 10–15 business days), and keep all follow-up communications logged.

Sample request wording (short)

Dear [Creditor name]

I received your adverse action notice dated [date]. Under ECOA/Regulation B and related guidance, I request a specific explanation of the principal reason(s) my application was denied and ask you to identify how that reason maps to my application or other data you used. If you relied on a credit report or score, please provide the consumer reporting agency name and a copy of the report and score factors. Please also provide (1) the model inputs and the factors used in my case, (2) any third-party data providers you relied on, and (3) the process for requesting a human reconsideration. Please respond in writing within 15 business days.

Sincerely,
[Your name]

Customize the letter to attach documents and to call out any known errors (for example, incorrect account numbers or misattributed transactions).

Common lender pushback — and how to reply

Expect two common replies: (1) the lender says the decision was driven by a black-box model and can't be explained, or (2) the lender refuses to share model code or training data claiming trade secrets. While some technical details and proprietary code may be withheld, regulators and legal commentators note that creditors are still required to provide specific principal reasons and, where possible, a mapping between the model’s outputs and the consumer-specific inputs used. If a lender refuses to provide a meaningful mapping or explanation, note that the CFPB has warned that compliance with adverse-action rules applies even when AI is used.

If the lender claims trade secrets, ask for:

• a redacted or high-level description of the model inputs used for your decision;
• a mapping table that ties the principal reason to named classes of data (e.g., "bank deposit patterns" or "merchant category transactions") rather than code;
• a confirmation of whether any protected-class proxies (race, national origin, etc.) or third-party surveillance signals were used.

Often an explanation that connects the principal reason to concrete, addressable actions (what you could change) is sufficient to permit a meaningful dispute or reconsideration, even if the lender won’t disclose full model architecture or training sets.

Appeals, disputes and escalation: next steps if the creditor won’t cooperate

1) Dispute inaccurate consumer-report items. If the adverse action relied on a consumer report and you find incorrect items, file a dispute with the reporting agency (CRA). The FCRA requires the CRA to investigate and the furnisher to verify or correct the information.

2) Request reconsideration and provide evidence. Re-submit supporting documents and ask the creditor to re-run a manual review with the corrected file or added information.

3) File a complaint with the CFPB and consider state enforcement channels. If the creditor refuses to provide a meaningful explanation or you suspect unlawful discrimination, submit a complaint at the CFPB’s complaint portal and include copies of the adverse action notice and your follow-up requests. Complaints can trigger supervisory attention or investigations by regulators and help build a public record.

4) Consider legal help. If you suspect discrimination or systemic failure to provide required notices, an attorney can advise on formal discovery requests, a state AG complaint, or litigation. Regulator guidance and legal commentary increasingly treat inadequate adverse-action explanations in AI-driven decisions as an enforcement risk for creditors.

Checklist — what to send and keep

• Copy of the adverse action notice and any pre-adverse communications.
• Written request for specific principal reasons and model evidence (keep a copy and send tracked mail or email).
• Request to the CRA for a free copy of your consumer report and score (if applicable).
• Records that support reconsideration: paystubs, bank statements, letters of explanation, ID verification.
• Log of dates, names, and phone calls with customer service or underwriting.

Keep all correspondence organized — if you eventually escalate to the CFPB or an attorney, well-documented records materially strengthen your position.

Bottom line: Regulators have made it clear that using AI doesn’t free lenders from the duty to tell consumers the real, principal reasons for adverse credit decisions. Ask for a plain-English mapping from the model’s principal reason to the data points in your file, get your consumer report and score details, request a human review, and escalate to the CFPB or a lawyer if the lender refuses to provide a useful, specific explanation.