Inside US banks' decades-long algorithmic fairness programs: new study

A new study, led by Emily Black (Carnegie Mellon University) and co-authors, offers the first empirical look at how the longest-running algorithmic fairness programs actually work in practice. Drawing on 35 semi-structured interviews with participants across the fair lending ecosystem—including compliance officers, regulators, auditors, and civil rights advocates—the researchers map out the policies, processes, and practices used by U.S. financial institutions to comply with fair lending laws. Unlike many recent AI fairness proposals, these programs have been running for decades, yet remarkably little was known about their day-to-day operations until now.

The study finds that while financial institutions maintain a 'floor' of fairness practices largely absent in other domains, the specifics of how firms test for discrimination and search for less discriminatory algorithms vary widely. Regulatory supervision via fair lending examinations has been the most effective driver of compliance work, but success often hinges on navigating competing business incentives, perceived legal tensions, and regulatory uncertainty. The paper will be presented at the 2026 ACM Conference on Fairness, Accountability, and Transparency (FAccT). Its findings underscore the unique role of supervisory authority in fostering fair lending practices—a design feature almost completely absent from recent policy proposals for tackling algorithmic discrimination in other sectors.