On September 19, 2023, the Consumer Financial Protection Bureau (CFPB) issued Consumer Financial Protection Circular 2023-03, addressing how lenders must handle adverse-action notices when they use artificial intelligence and other complex algorithms in credit decisions. Under the Equal Credit Opportunity Act and its implementing Regulation B, a creditor that denies credit, or offers worse terms, must tell the applicant the specific principal reasons for that decision.
The circular’s core holding is that this duty does not relax just because the decision came from a complex model. Lenders cannot simply pick the closest item from the CFPB’s list of sample reasons in Regulation B if that item does not actually describe what the model weighed. If a model relies on unusual or non-intuitive data - for example data harvested from consumer surveillance or factors a borrower would never expect to affect their creditworthiness - the lender must still disclose the actual principal reasons accurately and specifically. The circular gave the example that telling an applicant their “income was insufficient” would not suffice if the real driver was, say, the applicant’s profession.
The guidance is significant because many AI underwriting models are opaque, and a common industry hope was that regulators would accept generic, checklist-style explanations. The CFPB said no: the legal obligation to give a real reason is the same whether a human or an algorithm made the call, which in practice pressures lenders to use models they can actually explain.
Why business readers should care: this is one of the clearest statements that existing consumer-protection law already binds AI systems. A model you cannot explain is a model whose denials you may not be able to lawfully justify - a direct constraint on deploying black-box AI in lending.