On May 26, 2022, the Consumer Financial Protection Bureau issued a Circular on consumer financial protection stipulating that creditors using algorithmic tools in credit decisions must provide “statements of specific reasons to applicants against whom adverse action is taken” in accordance with the ECOA and Regulation B. The CFPB previously stated that the circulars are policy statements intended to “provide guidance to other agencies with consumer financial protection responsibilities on how the CFPB intends to administer federal consumer finance law”. The circular in question posits that certain complex algorithms constitute an uninterpretable “black box”, which makes it difficult, if not impossible, to identify with precision the specific reasons for the refusal of credit or the taking of other adverse measures. The CFPB concluded that “[a] The creditor cannot justify non-compliance with ECOA and Regulation B requirements based on the simple fact that the technology it uses to assess claims is too complicated or opaque to understand. »
This latest circular follows a proposal by the CFPB regarding the revision of the AI used in automated valuation models (“AVM”). As we noted in our previous post on this topic, the CFPB has stated that certain algorithmic systems could potentially violate the ECOA and its regulations (“Regulation B”). In this preview of proposals Regarding data capture, the CFPB recognized that some machine learning algorithms can often be too “opaque” for auditing. The CFPB further hypothesized that algorithmic models “may reproduce historical patterns of discrimination or introduce new forms of discrimination due to the way a model is designed, implemented, and used.”
In accordance with Regulation B, a statement of the reasons for the adverse action taken “must be specific and state the primary reason(s) for the adverse action. Statements that the adverse action was based on the creditor’s internal standards or policies or that the plaintiff, co-plaintiff or similar party did not achieve a qualifying score on the creditor’s credit reporting system are insufficient . In the circular, the CFPB reiterated that by using model disclosure forms, “if the reasons listed on the forms are not the factors actually used, a creditor will not satisfy the notification requirement by simply checking the closest identifiable factor listed”. In other related advisory opinion, the CFPB earlier this month also asserted that the provisions of ECOA and Reg B apply not only to applicants for credit, but also to those who have already received credit. This position echoes the Bureau’s previous position amicus brief on the same subject filed in John Fralish vs. Bank of Am., NA, ns. 21-2846(L), 21-2999 (7th Cir.). As a result, the CFPB says the ECOA requires lenders to provide “notices of adverse action” to borrowers with existing credit. For example, the CFPB says the ECOA prohibits lenders from lowering the credit limit on certain borrowers’ accounts or subjecting certain borrowers to more aggressive collection practices on a prohibited basis, such as race.
The most recent CFPB circular signals a less favorable view of AI technology compared to previous Bureau statements. In a July 2020 blog post, the CFPB highlighted the consumer benefits of using AI or machine learning in credit underwriting, noting that it “has the potential to expand access to credit by enabling lenders to assess the creditworthiness of some of the millions of consumers who are not rated using traditional underwriting techniques.” The CFPB also acknowledged that uncertainty regarding the existing regulatory framework may slow the adoption of such technology.At the time, the CFPB indicated that the ECOA maintained a level of “flexibility” and believed that ‘”A creditor need not describe how or why a disclosed factor adversely affected a claim…or, for credit reporting systems, how the factor relates to creditworthiness. In that previous article, the CFPB concluded that “a creditor may disclose a reason for denial even though the relationship between that disclosed factor and the prediction of creditworthiness may not be clear to the applicant. This flexibility may be helpful to creditors when issuing adverse action notices based on AI models where the variables and the main reasons are known, but which may be based on non-intuitive relationships.This message also highlighted the policy of Bureau’s no-action letter and compliance assistance sandbox policy as tools to help provide a haven for AI development. recent statement, the CFPB has criticized these programs as ineffective and it appears that these programs are no longer a priority for the Bureau. Similarly, this previous blog post now includes a disclaimer stating that it “provides an incomplete description of the ECOA Adverse Action Notice and Regulation B requirements, which apply from the same way to all credit decisions, regardless of the technology used to make them.ECOA and Regulation B do not allow creditors to use technology for which they cannot provide specific reasons for actions The disclaimer directs readers to the recent CFPB circular, which provides more information. This latest update makes it clear that the CFPB will look closely at the underpinnings of systems using such technology and demand explanations. detailed for their findings.[View source.]