From a glance at the paper (https://arxiv.org/pdf/2605.27371) it looks like they're using self-reported ethnicity.

"We only attempt to identify adverse impact based on self-reported racial information and do not impute data, noting that we do not have access to features like applicant name. 62.35% of all applicants in the data do not self-report race as belonging to one of the four racial groups we study."