AI Resume Screening & Disparate Impact Audits

Yes. Under Title VII of the Civil Rights Act of 1964 and EEOC formal guidance, an employer cannot delegate its non-discrimination obligations to a third-party software vendor. If an automated employment decision tool (AEDT) creates an adverse disparate impact under the Uniform Guidelines on Employee Selection Procedures (29 C.F.R. Part 1607), the employer is strictly liable as the principal. Furthermore, under the landmark ruling in Mobley v. Workday, Inc. (N.D. Cal. 2024), ATS software vendors themselves may be sued as statutory agents, but vendor indemnification clauses rarely cover statutory back pay, punitive damages, or class action settlements.

Under 29 C.F.R. § 1607.4(D), a selection rate for any race, sex, or ethnic group which is less than four-fifths (4/5 or 80%) of the rate for the group with the highest selection rate will generally be regarded by federal enforcement agencies as evidence of adverse impact. In AI resume screening, if male applicants advance at a rate of 40% and female applicants advance at a rate of 25%, the impact ratio is 62.5% (25/40), which falls substantially below 80% and establishes a prima facie case of disparate impact requiring statistical business-necessity validation.

NYC Local Law 144 mandates that employers using Automated Employment Decision Tools (AEDTs) to screen candidates residing in NYC must: (1) subject the tool to an independent annual bias audit conducted by an impartial auditor; (2) publicly publish a summary of the bias audit results and selection/impact ratios on their website; (3) provide candidates at least 10 business days advance notice before using the AEDT; and (4) allow candidates to request an alternative screening process or reasonable accommodation. Violations trigger civil penalties of up to $1,500 per violation per day.

Even when protected demographic characteristics (race, gender, age, disability) are excluded from the training data, machine learning models frequently latch onto proxy variables that correlate heavily with protected classes. Examples include high school graduation year or athletic participation (proxies for age), participation in women's sports or historically Black colleges and universities (proxies for sex and race), and gaps in employment history (proxies for pregnancy, caregiving, or medical disability). If an algorithm weights these features negatively, it produces an unlawful disparate impact.

An independent bias audit must be conducted by an objective, third-party data scientist or psychometric firm that has no commercial interest in the vendor's software and was not involved in developing, distributing, or licensing the tool. The auditor must calculate selection rates and impact ratios across all protected EEO-1 categories (race, ethnicity, sex, and intersectional identities) based on historical applicant data or synthetic test datasets that reflect actual deployment conditions.

Configuring AI algorithms to automatically downgrade or reject resumes showing career gaps poses severe disparate impact risks under Title VII, the Pregnancy Discrimination Act (PDA), and the Americans with Disabilities Act (ADA). Career breaks frequently reflect childbearing, infant care, eldercare (which disproportionately affect female applicants), or medical treatment for episodic disabilities. Employers must ensure algorithms evaluate qualifying competencies rather than penalizing chronological employment continuity.

Under Griggs v. Duke Power Co. and Title VII (42 U.S.C. § 2000e-2(k)), once an applicant proves disparate impact, the employer can only escape liability by proving that the screening practice is 'job-related for the position in question and consistent with business necessity' through formal psychometric validation (criterion, content, or construct validity under 29 C.F.R. Part 1607). Even then, the employer loses if the plaintiff demonstrates that an alternative selection practice with less adverse impact was readily available.

In Mobley v. Workday, Inc. (N.D. Cal. July 2024), the federal district court denied Workday's motion to dismiss, holding that third-party AI ATS platform providers can be held liable as 'employment agencies' or 'agents' under Title VII, ADEA, and ADA if employers delegate traditional hiring screening functions to the algorithmic tool. This precedent prevents employers from hiding behind vendor proprietary shields and triggers joint enterprise exposure during discovery.

Yes. The Office of Federal Contract Compliance Programs (OFCCP) issued definitive guidance stating that federal supply and service contractors using AI, machine learning, and automated screening systems must maintain detailed validation records, retain all algorithmic scoring outputs for a minimum of two years, and proactively audit hiring funnels under Executive Order 11246. Failure to produce algorithmic validation during a corporate management compliance review results in contract cancellation and debarment.

Under NYC Local Law 144, California proposed AI employment regulations, and EU AI Act extraterritorial standards, employers must provide written disclosure prior to application submission. The notice must specify: (1) that an automated employment decision tool will be used to evaluate qualifications; (2) the specific job qualifications and characteristics the tool evaluates; (3) the applicant's right to request an accommodation or alternative human-led assessment; and (4) instructions on how to access the employer's published annual bias audit.