Employment decisions made by algorithm rather than human judgment are increasingly the norm. Applicant tracking systems, automated resume screeners, AI-powered assessments, and machine learning ranking tools now control access to hundreds of millions of job opportunities. What has not changed—and what plaintiffs' practitioners need to understand clearly—is that the analytical framework for challenging these tools under Title VII's disparate impact theory is structurally intact, even as the political environment around federal enforcement shifts. The plaintiff who constructs a rigorous statistical case under the Griggs framework has a viable path forward regardless of the current administration's enforcement posture.
The Foundational Framework: Griggs v. Duke Power Co.
Griggs v. Duke Power Co., 401 U.S. 424 (1971), established the principle on which all disparate impact litigation rests. Duke Power required a high school diploma and passage of intelligence tests for transfer to certain departments. The tests had nothing to do with the ability to perform the jobs in question. Black applicants failed the tests at substantially higher rates than white applicants. The Supreme Court held, unanimously, that Title VII's prohibition on discrimination based on race is not confined to intentional discrimination: "Congress directed the thrust of the Act to the consequences of employment practices, not simply the motivation." Griggs, 401 U.S. at 432.
The Griggs rule: where an employment practice, neutral on its face, operates to exclude a protected class at a substantially higher rate than others, the employer bears the burden of demonstrating that the practice is "a reasonable measure of job performance" and "manifest[ly] relat[ed] to the employment in question." Id. at 431–32.
Statutory Codification: 42 U.S.C. § 2000e-2(k)
After the Supreme Court narrowed the Griggs doctrine in Wards Cove Packing Co. v. Atonio, 490 U.S. 642 (1989), Congress restored and codified the original framework in the Civil Rights Act of 1991. 42 U.S.C. § 2000e-2(k) now provides:
"(A) An unlawful employment practice based on disparate impact is established under this subchapter only if— (i) a complaining party demonstrates that a respondent uses a particular employment practice that causes a disparate impact on the basis of race, color, religion, sex, or national origin and the respondent fails to demonstrate that the challenged practice is job related for the position in question and consistent with business necessity; or (ii) the complaining party makes the demonstration described in subparagraph (C) with respect to an alternative employment practice and the respondent refuses to adopt such alternative employment practice."
This three-step burden-shifting framework is the operative statute for algorithmic discrimination claims:
- Plaintiff's prima facie case: Identify a specific, facially neutral employment practice and demonstrate statistically that it causes a disparate impact on a protected class.
- Employer's business justification: If the plaintiff makes the prima facie showing, the burden shifts to the employer to demonstrate the practice is job-related and consistent with business necessity.
- Alternative practice: Even if the employer carries its burden, the plaintiff may prevail by demonstrating that an equally effective alternative practice would produce less discriminatory impact and the employer refused to adopt it.
Mapping the Framework onto Algorithmic Tools
The application of this framework to AI hiring tools involves three analytical challenges that do not exist in traditional employment discrimination cases.
Challenge One: Identifying the "Specific Employment Practice"
Section 2000e-2(k)(1)(B)(i) requires the plaintiff to identify the specific challenged practice—not the overall decision-making process as a monolith. With algorithmic tools, this requirement creates difficulty: the "practice" that produces disparate impact may be a combination of training data, feature selection, model architecture, weighting schemes, and output thresholds that are opaque and proprietary.
Plaintiffs may argue that the entire algorithmic screening process is a single employment practice when "the elements of a respondent's decisionmaking process are not capable of separation for analysis." § 2000e-2(k)(1)(B)(i). This is the most important pleading strategy in algorithmic cases: allege facts establishing that the AI system functions as an integrated, indivisible decision-making process, making it impossible to isolate which specific component causes the disparity. Where this is plausibly alleged—and it frequently is, given the black-box nature of proprietary AI models—the process can be challenged as a whole.
Challenge Two: Statistical Proof of Causation
Disparate impact requires a demonstration of "significant" statistical disparity attributable to the challenged practice. Courts apply various statistical tests, but the EEOC's four-fifths (80%) rule remains the most commonly referenced benchmark: adverse impact is generally indicated when a protected group's selection rate is less than 80% of the selection rate of the group with the highest rate.
In algorithmic hiring cases, the statistical analysis requires: (a) identification of the relevant labor market and applicant pool; (b) selection rate data for each protected class; (c) demonstration that the algorithmic tool—not pre-application filtering or other factors—produced the disparity; and (d) statistical significance testing (typically using chi-square or standard-deviation analysis) to confirm the disparity is not attributable to chance.
Obtaining the data necessary for this analysis from a defendant employer or AI vendor requires targeted discovery. Build your requests around: applicant-level demographic data (or proxies if demographic data is not collected), algorithmic decision outputs with timestamps, training data documentation, and bias-testing records.
Challenge Three: Who Is the Employer?
Traditional disparate impact cases proceed against the employer. When the discriminatory tool is deployed by a third-party AI vendor, the question becomes whether the vendor is itself subject to Title VII and, if so, under what theory.
Mobley v. Workday, Inc., Case No. 3:23-cv-00770-RFL (N.D. Cal.), has been the central litigation testing these questions. Judge Rita Lin's July 12, 2024 ruling held that Workday could face direct Title VII liability as an "agent" of the employer where the employer had delegated traditional hiring functions—including candidate rejection and advancement decisions—to Workday's algorithmic system. The court rejected Workday's characterization as a mere passive software provider, reasoning that "drawing an artificial distinction between software decisionmakers and human decisionmakers would potentially gut anti-discrimination laws in the modern era." In May 2025, the court conditionally certified an ADEA collective action, finding common questions suitable for collective treatment.
Mobley establishes that AI vendors who exercise genuine, non-ministerial decision-making authority in the hiring process may be directly liable as agents under Title VII, the ADA, and the ADEA. Employers who deploy such tools remain liable as the direct employer. The practical result is that both can be named as defendants in a properly pled complaint.
EEOC Guidance on AI and Algorithmic Tools
The EEOC has issued guidance documents on employer use of AI and algorithmic decision-making tools in employment. In its 2023 guidance on artificial intelligence and the ADA, the EEOC emphasized that standard anti-discrimination principles apply to automated tools regardless of the technological mechanism. For Title VII's disparate impact framework, the EEOC has confirmed that the four-fifths rule applies to algorithmic selection procedures, and that an employer who uses an AI tool developed by a third party remains responsible for ensuring the tool does not produce unlawful disparate impact. The EEOC's amicus brief in Mobley is particularly instructive, endorsing the "agent" and "indirect employer" theories of AI vendor liability.
Note: the current administration has moved to reduce disparate impact enforcement at the federal level through executive action (April 2025). This affects federal agency enforcement priorities—not the viability of private Title VII disparate impact claims, which remain statutory rights cognizable in federal court by private plaintiffs.
Business Justification and Its Limits
If the plaintiff establishes prima facie disparate impact, the employer must demonstrate job-relatedness and business necessity. For AI tools, this requires the employer to produce: (a) validation studies showing the tool predicts job performance; (b) evidence that the specific criteria used are actually related to job requirements; and (c) documentation of bias testing conducted before and after deployment.
The "consistent with business necessity" standard is not met by showing that the tool is commercially available, that competitors use it, or that it is cost-efficient. Convenience and efficiency are not "business necessity." Griggs, 401 U.S. at 431 ("[t]he touchstone is business necessity"). An employer who deploys an AI tool without conducting validation testing or examining whether the tool disparately impacts protected classes will struggle to meet this burden.
Alternative Business Practices
Even where the employer carries its business necessity burden, the plaintiff can prevail by identifying a less discriminatory alternative that would equally serve the employer's legitimate needs. In the algorithmic context, this might include: using a modified algorithm trained on less biased data; applying human review at decision points where protected-class disparities appear; using blind resume screening; or replacing the challenged assessment with a validated, less discriminatory tool.
Expert testimony from industrial-organizational psychologists and AI auditing specialists is essential to making this showing. The plaintiff's expert need not design the alternative tool—only demonstrate, with sufficient specificity and statistical support, that a viable alternative exists that would materially reduce the disparate impact without sacrificing the legitimate business objective.
Practice Notes
At the pleading stage: Allege the algorithmic tool as a unified employment practice (to invoke the "not capable of separation" exception); allege specific demographic outcome disparities based on available information or information obtainable through discovery; and name both the employer and the AI vendor (under agency theory) as defendants.
Statistical evidence: Retain a quantitative expert early. Disparate impact cases live and die by the quality of the statistical analysis. Determine in early discovery whether the employer collected demographic data on applicants—if not, argue that the failure to collect such data supports an inference of intentional disregard for disparate impact.
EEOC charge filing: Title VII disparate impact claims require a prior EEOC charge before suit. The charge must be filed within 180 days of the discriminatory practice (or 300 days in a "deferral" state with a state agency). Given that algorithmic screening is a continuing practice, the continuing-violation doctrine may allow charges to relate back to earlier discriminatory applications—but do not rely on this without careful analysis.
Discovery of the algorithm: Proprietary AI systems will be vigorously defended. File targeted discovery requests for model documentation, training data demographics, bias testing records, and internal communications about disparate impact analysis. Anticipate trade-secret objections and prepare for a Rule 26(c) protective order negotiation.
Talk to Yates Anderson
If you are litigating a matter in this area — or weighing whether to — the working analysis above only goes so far. Request a case evaluation and a Yates Anderson attorney will respond within one business day.
Informational only. Not legal advice. No attorney-client relationship is created by reading this post. Consult a licensed attorney in your jurisdiction.