The CFPB recently issued a new no-action letter (NAL) to Upstart Network, Inc. regarding its automated model for making underwriting and pricing decisions on applications by consumers for unsecured, closed-end loans.  Upstart uses artificial intelligence techniques and alternative data in its model.

The new NAL is essentially a renewal of the NAL issued to Upstart in September 2017, which was the first NAL issued by the CFPB.  The first NAL had a three-year term that was originally set to expire on September 11, 2020.  The Bureau extended the first NAL’s expiration date until December 1, 2020.

The new NAL, which has a 36-month term, was issued under the Bureau’s revised NAL Policy issued in September 2019.  It provides that unless the NAL is terminated by the Bureau pursuant to the revised NAL Policy, the Bureau will not make supervisory findings or bring a supervisory or enforcement action against Upstart under (1) the ECOA or Regulation B for discriminating against an applicant on a prohibited basis or for discouraging an applicant on a prohibited basis, or (2) the Bureau’s UDAAP authority.

In its application for the new NAL, Upstart agreed to significantly expand its reporting obligations to the Bureau and engage in additional fair lending testing beyond the testing and reporting under the initial NAL.  The expanded reporting and additional testing is set forth in a Model Risk Assessment Plan (MRAP) that Upstart agreed to enter into with the Bureau.  The new NAL is conditioned on implementation of the MRAP.  While the new NAL notes that the MRAP contains confidential information that the Bureau is prohibited from disclosing, it lists certain of the MRAP’s requirements.  Upstart is required to:

• Notify the Bureau of significant changes to Upstart’s model before implementation
• Provide the Bureau with model documentation on a periodic basis, including a technical report describing certain aspects of each component of Upstart’s model and performance monitoring reports that evaluate how Upstart’s customer population and model performance change over time
• Test Upstart’s model and/or variables or groups of variables on a periodic basis for adverse impact and predictive accuracy by groups, and provide results to the Bureau
• Research approaches that may produce less discriminatory alternative models that meet legitimate business needs
• In addition to fair lending testing, conduct periodic access to credit testing to determine how Upstart’s model compares to other credit models in enabling credit access and provide results to the Bureau
• Provide the Bureau access to the software code used to implement the MRAP