On September 19, the Consumer Financial Protection Bureau (CFPB) issued Circular 2023-03
, which provides guidance as to how lenders must explain denials of applications of credit when the underwriting is based on artificial intelligence (AI) or complex credit models. The upshot of the guidance is that a lender that denies an application for credit based on AI or similar predictive decision-making technologies may not rely on the use of a standard “checkbox” adverse action notice form, if the use of the form would not “accurately describe the factors actually considered or scored by a creditor.” Instead, the lender must provide the specific information that led to the denial. The CFPB used the example that an explanation that the applicant had “insufficient projected income” or “income insufficient for amount of credit requested” would be inadequate if the actual basis for the denial was an income estimation based on the applicant’s profession.
Circular 2023-03 follows two 2022 publications from the CFPB (Circular 2022-03 and a May 2022 Advisory Opinion) that expressed similar concerns that lenders may be attempting to avoid their obligations under the Equal Credit Opportunity Act (ECOA) and Regulation B to avoid unlawful discrimination and explain bases for the denial of credit applications by the use of more complicated “black box” underwriting technology.
The CFPB’s repeated warnings have come in an environment where more companies are leveraging technology to use new methods of underwriting that can offer better predictive results than traditional credit scores. This new wave of “alternative” underwriting provides new means for extending credit to applicants who would not be eligible under traditional credit-score-based underwriting – but critics contend that it also introduces new methods for lenders to engage in discrimination through the use of informational proxies.
For now, the CFPB’s publications serve more as a warning shot than a concrete set of principles governing the use of AI credit underwriting. The key takeaway is that lenders need to have a much deeper understanding of the technologies that they’re using rather than just a surface-level familiarity, and adverse action notices need to provide as specific information as possible. The next frontier will be understanding the disparate impacts of AI underwriting and preparing for an anticipated regulatory response.