On July 10, 2025, the Massachusetts Attorney General’s Office (AGO) reached a $2.5 million settlement with a private student loan company to resolve allegations that its lending practices violated various consumer protection and fair lending laws. These practices included the use of artificial intelligence (AI) models and manual underwriting that the AGO alleged led to disparate impact harm to Black, Hispanic, and non-citizen applicants and borrowers. The settlement was the latest example of state enforcement activity in the wake of scaled-back federal enforcement under the current administration.
Specifically, the AGO alleged that the student loan company:
- Used the cohort default rate (CDR), a dataset produced by the U.S. Department of Education, as a variable in its AI models, which disproportionality hurt the approval rates and resulted in less desirable loan terms for Black and Hispanic applicants;
- Employed “knockout rules” that automatically denied applicants during the “prescreen decline” underwriting stage if they did not have at least a green card, without any determination of the applicants’ creditworthiness or ability to repay;
- Generated inaccurate and non-specific reasons in adverse action notices in violation of the Equal Credit Opportunity Act (ECOA) and Regulation B;
- Allowed underwriters to manually override AI model decisions without proper controls; and
- Failed to safeguard against disparate outcomes by testing and implementing adequate policies and procedures.
As we wrote about previously, President Trump signed an executive order titled “Restoring Equality of Opportunity and Meritocracy” on April 23, 2025, which eliminated the use of disparate-impact liability in various contexts at the federal level. There has since been speculation that state attorneys general offices, such as the AGO in this matter, would step in and further their use of disparate impact claims. The AGO’s settlement demonstrates states’ continued use of disparate impact liability under federal and state civil rights laws, despite the federal government’s intent for the executive order to be far-reaching. Private plaintiffs can also continue to bring disparate impact claims of discrimination in court. In light of the AGO’s settlement, lenders are therefore advised to continue their disparate impact risk management and to avoid abrupt changes in their fair lending policies.
Key Takeaways
Based upon the settlement, best practice takeaways for lenders include:
- Establishing clear written policies for approving underwriting exceptions to AI model decisions, while also making sure that any exceptions are appropriately documented and retained.
- Prohibiting the use of discretionary underwriter judgment without approval from senior underwriting officers or the credit committee, or that deviates from the lender’s written policies.
- Conducting annual fair lending testing of AI models, including additional testing for certain events such as model updates or the creation of new models.
- Conducting periodic fair lending analysis, including matched pairs testing, to determine whether there are instances where two applicants with similar financial circumstances are subject to different outcomes.
- Ensuring that all adverse action notices indicate specific and accurate reasons for the adverse action.
- Monitoring and testing to ensure that certain weighted variables used by AI models do not cause disparate impact harm to protected class groups.
- Refraining from using variables in AI models that automatically disqualify groups of applicants without having the applicant go through the full underwriting process.
