CFPB Issues Supervisory Highlights on Advanced Technologies

The CFPB recently issued a special edition of Supervisory Highlights focused on select examinations of institutions that use credit scoring models, including models built with advanced technology commonly marketed as artificial intelligence and machine learning (AI/ML) technology, when making credit decisions.  The CFPB provided examples of recent institutions that used such models in the underwriting and pricing of applications and in making credit decisions. 

Examiners found disparities in underwriting and pricing outcomes for Black or African American and Hispanic applicants, as well as deficient compliance management systems.  Examiners also determined that the institutions did not maintain fair lending controls capable of evaluating and addressing the risks associated with their credit scoring models to ensure compliance with ECOA and Regulation B.

In assessing compliance with ECOA and Regulation B with respect to such models, examiners found that in some cases, institutions used models that used more than a thousand input variables.  Examiners identified risks associated with the use of such a large number of input variables, including difficulties in being able to effectively monitor whether any variables, individually or in combination, acted as a proxy for prohibited basis under ECOA.  Moreover, when evaluating models for disparate impact, institutions did not meaningfully identify and consider comparably accurate inputs with less discriminatory effects, nor did they adequately document the business need for the model inputs the institutions identified as contributing to prohibited basis disparities. 

Examiners also found that the institutions did not sufficiently ensure compliance with adverse action notice requirements, including how they selected the reasons given in adverse action notices when the adverse action was based on the model score. Examiners also found that the institutions had not validated that their processes for selecting reasons produced accurate results.