TL;DR: This page documents AI hiring bias examples under five evidence labels: finding, settlement or consent decree, allegation, reporting and study. One example ended in an EEOC consent decree (iTutorGroup, $365,000, 2023). In the AI-vendor court cases listed here, no court has found a vendor liable; those cases are allegations. Lawsuits are tracked on the tracker page.
General information, not legal advice. Have counsel confirm how this applies to you.
The documented examples, each with its label:
- Amazon's experimental recruiting tool (reported 2018): reporting
- iTutorGroup age screening (EEOC consent decree, 2023): settlement or consent decree
- University of Washington resume-screening study (2024): study
- Mobley v. Workday (allegations, pending): allegation
Updated 28 October 2026. Next review: 28 November 2026. The page is reviewed monthly while Mobley v. Workday is active (class-certification hearing set for 9 March 2027), on any ruling in Mobley v. Workday or Harper v. Sirius XM, and quarterly otherwise.
Key takeaways
- Each documented example carries one evidence label, because an allegation, a settlement, a media report and a research study support different conclusions.
- The EEOC's iTutorGroup case, settled by consent decree in September 2023 for $365,000, alleged a hard-coded age cut-off in application software, not a learned model (EEOC release, 11 September 2023).
- No court has issued a liability finding in the AI-vendor cases listed on this page; they are pending allegations.
- The University of Washington study tested language models ranking real-world resumes in a research setup that approximates real-world systems, not live hiring decisions (UW news, 31 October 2024).
- A Local Law 144 bias audit is required to test only sex, race/ethnicity and their intersections; age and disability, central to several of these cases, are absent from its required calculations (6 RCNY § 5-301(b)–(c), April 2023).
An allegation in a complaint, a consent decree, a news report and a research study support four different conclusions, and they are easily listed side by side as though they carried the same weight. This page labels every example and says what each one does not show. One example here ended in a $365,000 consent decree; the Amazon example has no court record at all.
The labels matter most to employers that rely on audits under NYC's bias audit law, because what a case establishes determines what an audit can be expected to catch. Pending lawsuits change as dockets move, so they are kept on a separate tracker of AI hiring cases; this page carries short labeled rows for them.
How should you read these examples? The evidence labels
Every example carries exactly one of five labels:
- FINDING: a court or agency decision.
- SETTLEMENT or CONSENT DECREE: a resolution reached with an agency or a private party.
- ALLEGATION: a complaint, charge or class claim that has not been proven.
- REPORTING: journalism that cites its sources.
- STUDY: research, which does not involve real hiring decisions.
No hypotheticals are presented as cases. An example that cannot be sourced to a court, an agency or a credible publication stays out of the page.
Has anyone been fined for AI hiring bias? One example here ended in an EEOC consent decree, with payment to the rejected applicants (iTutorGroup, $365,000). In the AI-vendor cases listed here, no court has issued a liability finding. As of 28 October 2026, the Mobley v. Workday docket shows no such finding, and the Harper v. Sirius XM docket shows a judgment for the defendant.
The evidence ledger: every example, its label and what it does not show
The ledger comes first because it answers the main question: how much weight does each example carry?
| Example | Label | Court and docket number, or source | Date | What this does NOT show |
|---|---|---|---|---|
| Amazon experimental recruiting tool | REPORTING | Reuters, via the CNBC syndicated copy | 10 Oct 2018 | No court record. Anonymous sources. An internal experiment, not a deployed hiring decision system. Not a finding about any current product. |
| iTutorGroup | SETTLEMENT / CONSENT DECREE | EEOC v. iTutorGroup, Inc., et al., No. 1:22-cv-02565 (E.D.N.Y.); EEOC release | 11 Sep 2023 | A learned model. The alleged mechanism was a hard-coded age cut-off. Whether the decree contains a denial of liability is not stated in the EEOC release cited here. |
| Mobley v. Workday, Inc. | ALLEGATION (pending) | N.D. Cal., No. 3:23-cv-00770; CourtListener docket | Filed 21 Feb 2023 | Any finding of liability. The claims are allegations, and the case is pending. |
| Harper v. Sirius XM Radio, LLC | ALLEGATION | E.D. Mich., No. 2:25-cv-12403; CourtListener docket; Fisher Phillips, 15 Aug 2025 | Filed 4 Aug 2025 | Any finding of liability. The court granted Sirius XM's motion for judgment on 30 Sep 2026, and the plaintiff filed a notice of appeal on 27 Oct 2026. |
| University of Washington resume-screening study | STUDY | Wilson and Caliskan, AIES 2024 | Published 16 Oct 2024 | Real hiring decisions. Any claim about a hiring product. |
| Baker v. CVS Health Corporation | SETTLEMENT (statute claim, not a bias finding) | D. Mass.; New England Biz Law Update, 23 Feb 2024; Legal Dive, 24 Jul 2024 | Feb 2024; Jul 2024 | Discrimination. The claim was under a Massachusetts lie-detector statute. Settlement terms are undisclosed. |
| HireVue facial analysis | ALLEGATION in a complaint, then a product change | EPIC complaint to the FTC (2019); Fortune, 19 Jan 2021 | 2019; Jan 2021 | A finding of bias. This page does not state an FTC outcome. |
Status of pending cases changes; see the tracker linked above.
Which examples come from reporting and research?
Amazon's experimental recruiting tool (REPORTING)
Reuters reported on 10 October 2018 that Amazon had built an experimental tool to rate job candidates, and then scrapped it. The account rests on five people familiar with the effort. The text of the report is available through the CNBC syndicated copy (updated 11 October 2018).
According to the report:
- A team had been building resume-rating programs since 2014. By 2015, the company found the system was not rating software developer and other technical candidates in a gender-neutral way.
- The models were trained on 10 years of resumes, most of which came from men.
- The system penalized resumes that included the word "women's" and downgraded graduates of two all-women's colleges.
- Amazon edited the programs to neutralize those terms but could not guarantee the system would not sort candidates in other discriminatory ways. The team was disbanded by the start of 2017.
- Recruiters looked at the system's recommendations but, the sources said, never relied solely on them.
Four caveats apply. The sources are anonymous. There is no court record. The tool was an internal experiment, not a deployed hiring decision system. And the report is not a finding about any current product.
University of Washington resume-screening study (STUDY)
Researchers Wilson and Caliskan audited three large language models on how they ranked resumes for job descriptions when only the first names were swapped between names associated with White, Black, male or female people. According to the University of Washington (31 October 2024), the study used over 550 real resumes and 571 job listings. The models favored White-associated names 85% of the time and female-associated names 11% of the time, and male-associated names 52% of the time. The work was presented at the AAAI/ACM Conference on AI, Ethics, and Society and published on 16 October 2024 in its proceedings (AIES paper).
The study was a simulated retrieval test on language models. It did not involve real hiring decisions, and it makes no claim about any hiring product. Neither does this page: how AI interview software works covers that category, and nothing here is a fairness claim about any product, ZeaHire included.
Which examples have reached a court or an agency settlement?
iTutorGroup (SETTLEMENT / CONSENT DECREE)
EEOC v. iTutorGroup, Inc., et al., Civil Action No. 1:22-cv-02565 (E.D.N.Y.), was resolved by a consent decree that the EEOC announced on 11 September 2023. The EEOC brought the suit under the Age Discrimination in Employment Act. It alleged that the company programmed its application software to reject female applicants aged 55 or older and male applicants aged 60 or older automatically, and said more than 200 qualified U.S. applicants were affected.
The company agreed to pay $365,000 to the rejected applicants and to provide anti-discrimination training and a policy. It agreed to stop asking applicants for birth dates and to accept EEOC monitoring for at least five years. If it resumes U.S. operations, it must notify and interview the rejected applicants (EEOC release, 11 September 2023).
The nuance matters. The alleged mechanism was a hard-coded age cut-off, not a learned model. Law-firm commentary (Duane Morris, 11 August 2023) describes the case as the EEOC's first settlement involving AI hiring software, while the EEOC's own release refers to "application software." The Duane Morris post is dated a month before the EEOC's 11 September 2023 release, so this page uses the EEOC's date for the announcement. The EEOC release cited here does not say whether the decree contains a denial of liability.
Mobley v. Workday, Inc. (ALLEGATION; pending; no liability finding)
Mobley v. Workday, Inc., No. 3:23-cv-00770 (N.D. Cal.), was filed on 21 February 2023 and is before Judge Rita F. Lin. The plaintiffs allege that Workday's screening tools discriminated against applicants on the basis of age. On the CourtListener docket (last filing seen 10 November 2026), the court granted preliminary collective certification of an Age Discrimination in Employment Act collective on 16 May 2025. Plaintiffs filed a Fourth Amended Complaint on 10 July 2026. The court also granted in part and denied in part a motion to dismiss on 6 March 2026 and denied a motion for a certificate of appealability on 2 July 2026. The scheduling order set the class-certification motion for 14 September 2026, and the plaintiffs filed it that day. An opposition/response was filed on 10 November 2026, the response deadline; the plaintiffs' reply is due 8 December 2026 and the hearing is set for 9 March 2027 (docket read 11 November 2026).
These are allegations, not findings, and no court has found Workday liable. The full docket timeline and the court's rulings are on the Workday AI hiring lawsuit case page.
Harper v. Sirius XM Radio, LLC (ALLEGATION)
Harper v. Sirius XM Radio, LLC, No. 2:25-cv-12403 (E.D. Mich.), was filed on 4 August 2025 before Judge Terrence G. Berg (CourtListener). Fisher Phillips (15 August 2025) describes the claim as race discrimination through an AI hiring tool that relies on historical data. The case is listed here as an allegation. On the docket, the court granted Sirius XM's motion for judgment on 30 September 2026, and the plaintiff filed a notice of appeal to the 6th Circuit on 27 October 2026. Status of pending cases is maintained on the AI hiring lawsuits tracker.
Which legal risks are not bias findings?
Not every AI hiring lawsuit is a bias case. Listing the cases below as bias examples would be inaccurate, so they are separated here.
Baker v. CVS Health Corporation (statute claim). Baker v. CVS Health Corporation (D. Mass.) was a proposed class action against CVS Health Corporation and CVS Pharmacy, Inc. The New England Biz Law Update (23 February 2024) reported that the suit centered on a video interview technology called HireVue that CVS allegedly used, and that AI allegedly analyzed facial expressions, eye contact and voice. HireVue is named only as the alleged technology, not as a defendant. The plaintiff alleged this amounted to an unlawful lie-detector test under Massachusetts law. The same report says Judge Patti B. Saris denied dismissal of the notice claim. Legal Dive, 24 July 2024 reports a tentative settlement filed 17 July, with monetary terms not disclosed. The claim was brought under a statute. It was not a discrimination or bias finding.
HireVue and facial analysis (allegation, then a product change). In 2019, the Electronic Privacy Information Center filed a complaint asking the FTC to investigate HireVue's AI assessments. In January 2021, HireVue said it had removed the visual analysis component, which it said was discontinued in March 2020 because visual analysis no longer significantly added value (Fortune, 19 January 2021; SHRM). That is an allegation in a complaint followed by a product change. No finding of bias was made, and this page does not state an FTC outcome.
Kistler v. Eightfold AI Inc. (transparency claim). Kistler v. Eightfold AI Inc., No. 3:26-cv-01768 (N.D. Cal.), is a Fair Credit Reporting Act case before Judge Yvonne Gonzalez Rogers, so it is a transparency or consumer-reporting claim, not a discrimination claim. The CourtListener docket shows the federal case opened on 2 March 2026 after removal. As of the docket's 13 October 2026 update, Eightfold's motion to dismiss had no ruling. The tracker covers it.
Why does AI hiring bias happen? Five mechanisms, each tied to a case
The cases above point to five mechanisms. This table is our synthesis, and each row is tied to an example and graded by its evidence label.
| Mechanism | Case | Evidence grade |
|---|---|---|
| Learned from skewed historical outcomes | Amazon (reported) | Reporting |
| Hard-coded rule using a protected attribute | iTutorGroup (alleged by EEOC) | Consent decree |
| Name or group effects in language-model ranking | UW study | Study |
| Vendor as a screening gatekeeper | Mobley v. Workday | Allegation |
| Facial or voice analysis of interviews | HireVue (company changed product); CVS (statute claim) | Complaint / settlement, not bias finding |
Each row concerns how a tool was designed, trained or deployed, and the table makes no statement about any company's current product. The state AI hiring law tracker shows which jurisdictions now require audits, notices or impact assessments for these mechanisms.
What would a Local Law 144 bias audit have caught?
Start with what a bias audit tests. The categories the rule requires are sex, race/ethnicity and their intersections, and nothing else (6 RCNY § 5-301(b)–(c), April 2023). Read literally, an audit would test the sex dimension of an Amazon-type tool. It would not test age, which is central to iTutorGroup and to Mobley's age theory, and it would not test disability.
Other laws cover age and disability, and counsel decides how they apply to a given process. For the sex dimension, the audit's selection-rate arithmetic, explained in the four-fifths rule, is the calculation that would apply to a sex-based screen.
One dated line gives context on how audits are used in practice. A study of 391 New York City employers (Wright et al., June 2024) found that 18 had posted audit reports and 13 had posted notices.
The process of commissioning one is covered in how a bias audit is conducted.
What do these cases mean for how you hire?
The cases are not a verdict on AI in hiring. They point to three practical conclusions:
- Published selection data, and the ability to explain each decision, matter when a tool screens at volume.
- A human reviewer with authority and evidence matters when a tool's output feeds a decision.
- The legal context differs by jurisdiction, and the tracker is where to check it.
None of this is a claim that any tool, ZeaHire included, is free of bias. These cases show why published selection data and human review matter.
FAQ
Has anyone been fined for AI hiring bias?
One example here ended in an EEOC consent decree: iTutorGroup agreed in 2023 to pay $365,000 to rejected applicants. No court has issued a liability finding in the AI-vendor cases listed on this page.
Why do AI hiring tools become biased?
The mechanisms table above sets out four documented patterns: learning from skewed historical outcomes, hard-coded rules, name or group effects in language-model ranking, and vendors acting as screening gatekeepers. A fifth, facial or voice analysis of interviews, has produced a statute claim and a product change rather than a bias finding.
Is AI hiring bias illegal?
Discrimination in hiring is unlawful under federal, state and city law regardless of the tool used. Whether a given use violates a law is a question for a court or an agency. See the state AI employment laws tracker for the rules by jurisdiction.
Can a bias audit catch this?
Only within the categories the audit tests, which are sex and race/ethnicity. It does not test age or disability.
What is the Workday lawsuit about?
Mobley v. Workday, Inc. (N.D. Cal., No. 3:23-cv-00770) alleges that Workday's screening tools discriminated on the basis of age. The court has preliminarily certified an ADEA collective, and a class-certification motion is pending. The claims are allegations, and no court has found Workday liable. Full detail is on the Workday case page.
Was the Amazon tool ever used to evaluate candidates?
According to Reuters' sources, recruiters looked at the tool's recommendations but never relied solely on them. See the caveats in the Amazon entry above.
Are the CVS and HireVue cases bias cases?
No. The CVS case was a statute claim, and the HireVue matter was a complaint followed by a product change. See the section on legal risks that are not bias findings.
Next step
Look at how a ZeaHire evidence pack records the reasoning behind each question score: see an evidence pack in a demo. Your counsel decides what the law requires of your process.
Comments
No comments yet.
Comments are closed on this post.