Adverse action reason testing, fair lending analysis, servicemember checks and credit reporting dispute reviews are the strongest uses of AI agents in US consumer lending compliance. Compliance and fair lending counsel keep every conclusion and remediation. ECOA and Regulation B, CFPB supervision, SR 11-7, state rules and the Colorado AI act shape the work.
What does consumer lending compliance look like in the US?
Adverse action notices are a daily test. ECOA and Regulation B require lenders to give applicants specific reasons when credit is denied or offered on worse terms, including when a model or algorithm made or informed the decision. Generic reasons, or reasons that do not reflect what actually drove the outcome, are a recurring supervisory finding.
Fair lending analysis is the second load. Compliance tests whether approval rates, pricing and terms differ across protected groups, whether differences have legitimate business justification, and whether less discriminatory alternatives exist. Models make this both more important and more complex.
Then there are the many specific rules: truth-in-lending disclosures, accuracy of credit reporting and handling of disputes, protections for active-duty servicemembers and their dependants, debt collection conduct, and the general prohibition on unfair, deceptive or abusive acts or practices. Nonbank lenders also manage state licensing in every state where they lend.
Where can agents help?
- Adverse action reason testing (depth 1). The agent compares the reasons in each notice with the factors that actually drove the model's output and the underwriting file, and flags notices where reasons are generic or do not match.
- Fair lending analysis drafting (depth 2). The agent runs approved statistical tests across products and channels and drafts the analysis for fair lending counsel, citing data and methods.
- Servicemember checks (depth 1). The agent checks that covered borrowers were identified at origination and that rate limits and other protections were applied.
- Credit reporting dispute review (depth 1). The agent compares disputed items with account records and checks that investigations were completed and corrections furnished.
- Complaint theme analysis (depth 2). The agent groups complaints by product, channel and allegation, and drafts a summary that flags potential unfair or deceptive practices.
Adverse action testing is the clearest first use for lenders using machine learning models. The rule is explicit, and it applies however complex the model is. Our piece on the CFPB and AI agents in consumer finance goes deeper.
What stays with people?
Fair lending conclusions stay with counsel and compliance. Whether a disparity is justified, and whether a less discriminatory alternative must be adopted, are legal and business judgements with serious consequences. The agent runs the numbers; people decide.
Remediation, from re-issuing notices to refunding servicemembers, stays with named staff. So do responses to regulators and any self-identification of issues.
Credit decisions stay with the credit function and its models, which are governed separately. The compliance agent tests them after the fact. This is the principle of AI compliance agents as second line, never the last word.
Which US rules apply?
There is no single federal AI law. ECOA and Regulation B prohibit discrimination in credit and require specific reasons when credit is denied, including when a model is used. The CFPB supervises larger banks and many nonbank lenders for consumer financial law, including unfair, deceptive or abusive practices. The OCC, the Federal Reserve and the FDIC supervise banks, and their interagency model risk guidance, SR 11-7, applies to credit models and to agents that test them. State regulators, including NYDFS, license and examine nonbank lenders.
GLBA governs borrowers' financial privacy, and state privacy laws such as California's CCPA as amended by the CPRA may reach other personal data. Colorado has passed an AI act aimed at high-risk systems that make consequential decisions, which include lending decisions, so lenders active there should track it with counsel. See how AI agents are regulated in the United States for more.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
What does a strong first 90 days look like?
Month one: run adverse action reason testing at depth 1 on a sample of denials from each model-driven product. Bring results to fair lending counsel before anything else.
Month two: register the agent in the model inventory under SR 11-7 if you are a bank, or under your equivalent governance if not. Add servicemember checks and credit reporting dispute review at depth 1.
Month three: pilot fair lending analysis drafting at depth 2 on one product. Counsel reviews the method and every conclusion. The UK's outcome-based approach in FCA Consumer Duty and AI agents in lending, and the practice of monitoring customer outcomes with an agent, offer useful methods even though they are not US law.
Where Pryme fits
Pryme certifies adverse action testing and fair lending drafting as separate capabilities, each on the lender's own files through Attest. Agent drafts for counsel are held verdicts waiting for a named person. Every verdict is filed in an append-only record, which supports SR 11-7 monitoring and examiner requests.
Questions people ask
Can a compliance agent generate adverse action reasons?
It should test them, not generate them for live decisions. The reasons must reflect the actual factors in the credit decision, and producing them belongs to the credit process under the lender's governance.
Is a fair lending testing agent a model under SR 11-7?
It processes data to produce outputs that inform decisions, so treat it as one: document it, validate it and monitor it.
How does the US compare with the UK for consumer lending compliance?
The UK adds an outcomes-based Consumer Duty with board reporting. See our brief on compliance agents in UK consumer lending.