Hiring and lending: EEOC, Title VII, and the CFPB

Lesson 3 of 5 in Sectoral Enforcement: FTC, EEOC, CFPB, FDA, and Financial Regulators.

Employment. Three statutes from the civil-rights era do the work: Title VII (race, sex, religion, national origin), the ADA (disability), and the ADEA (age 40+). Two theories matter for algorithms. Disparate treatment: the tool intentionally uses a protected trait. Disparate impact: a facially neutral practice — say, a résumé-scoring model — disproportionately excludes a protected group and cannot be justified as job-related and consistent with business necessity. Disparate impact is the natural fit for AI, because models discriminate through proxies without anyone intending anything.

The ADA adds three AI-specific traps: screening out disabled candidates who could perform with accommodation (a video-interview model penalizing atypical speech), failing to offer accommodations in the assessment process itself, and questions that function as unlawful medical inquiries. The EEOC’s first algorithmic settlement was, fittingly, the simplest possible case: iTutorGroup (2023) paid $365,000 after its application software auto-rejected women 55+ and men 60+ — disparate treatment coded as a rule.

The enforcement climate shifted with the administration: the EEOC’s 2022–2023 AI technical-assistance documents (ADA guidance; Title VII selection-procedure guidance applying the four-fifths rule of thumb to algorithms) were withdrawn or removed in 2025, and federal disparate-impact enforcement was deprioritized by executive policy. Hold the distinction firmly: guidance is gone; the statutes are not. Title VII, the ADA, and the ADEA still support private suits — which is exactly what Mobley is — and state agencies and laws (Illinois HB 3773, NYC LL144) fill the federal gap. Detailed treatment of those state regimes lives in the state-laws module.

Credit. The CFPB’s contribution to AI law is one uncompromising idea: complexity is not a defense. The Equal Credit Opportunity Act and Regulation B require creditors taking adverse action — denial, limit cut, rate increase — to state the specific principal reasons. Two circulars applied this to machine learning: Circular 2022-03 — a creditor cannot use a model so complex it cannot generate accurate reasons (‘black box’ is a confession, not an excuse); Circular 2023-03 — reasons must be specific and accurate to the applicant, not plucked from the sample checklist.

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.