CFPB: The use of complex algorithms is not a legal defense against discrimination

The Consumer Financial Protection Bureau recently notified companies that under federal anti-discrimination laws still owe consumers an explanation of specific reasons for denying credit applications, even though they use complex algorithms to determine creditworthiness. This decision recalls both the agency’s continued focus on anti-discrimination enforcement as well as the sustainable responsibility companies using new technologies in interactions with consumers.

On May 26, the agency issued a circular confirming its position that notice of adverse creditor action requirements under the Equal Credit Opportunity Act apply when using artificial intelligence or other algorithm-based credit models, though the company says it doesn’t fully understand how the technology it uses to make those decisions works. Beyond denied credit applications, adverse actions may include closing or changing the terms of an existing credit account or denying a request to increase credit limits.

“Companies are not exempt from their legal responsibilities when they let a black box model make lending decisions,” CFPB Director Rohit Chopra said in a statement. Press release.

“The law gives every applicant the right to an accurate explanation if their credit application has been denied, and that right is not diminished simply because a company uses a complex algorithm that it does not understand.”

The circular comes after the CFPB announcement in mid-March that he would prioritize targeting unfair discrimination even if fair lending laws do not apply, citing prohibitions against unfair, deceptive and abusive practices under the Protection Act consumer finance (CFPA). In a move signaling closer collaboration with states, including state attorneys general, the CFPB is also empowerment of states to enforce CFPA provisions, recently issued a rule of interpretation clarifying that Section 1042 allows states to enforce any provision of the law. The interpretation rule notes that a CFPB action would not preempt a parallel state action. Further evidence of federal-state partnerships is the fact that the CFPB has entered into memorandums of understanding with nearly two dozen state attorneys general, all 50 states, the District of Columbia and Puerto Rico.

Ultimately, creditors and lenders are still liable under federal law if they don’t provide specific reasons for adverse actions, and a lack of understanding of how credit modeling technology works is not not a legal defense to non-compliance. More generally, companies operate in a regulatory environment state and federal level increasingly focused on protecting consumers from algorithmic discrimination. Companies would do well to review their algorithms for disparate treatment and disparate impact.

Sharon D. Cole