If your clients use artificial intelligence (“AI”) to screen job applicants—and the data suggest that most large employers do—they are probably not as protected as they think.
Previously, federal agencies treated disparate-impact liability as a legitimate tool by which to combat employment discrimination, allowing them to pursue claims where facially neutral practices produced statistically disproportionate outcomes for protected classes. That posture shifted abruptly in April 2025. President Trump’s executive order instituting a pullback on federal enforcement of disparate-impact claims has created a widespread impression that AI hiring risk is, for now, manageable.[1] That impression is wrong in two important ways, and the lawyers who help their clients understand both will be doing them a genuine service.
First, federal enforcement pullback does not touch private litigation. Second, executive order priorities reverse with administrations, even those of the same party. The conduct happening now generates liability that does not disappear when enforcement resumes. A new administration in 2029 will inherit accrued liability that it could suddenly decide to prioritize. This is a particularly large risk given the high level of public skepticism toward AI being used in the hiring process.
The Litigation Pipeline Is Already Building
The most prominent test case in this space is Mobley v. Workday, Inc., currently being litigated in the U.S. District Court for the Northern District of California.[2] The plaintiff, an African American man over forty, applied through Workday’s AI-powered screening platform to more than one hundred positions and was rejected from all of them, often within hours of applying. He sued Workday directly. The court has allowed the claims to proceed on an agent-liability theory, holding that drawing a legal distinction between human and software decision-makers would gut antidiscrimination law in the modern era. A nationwide collective action covering applicants over forty was conditionally certified in May 2025. In March 2026, the court rejected Workday’s argument that the Age Discrimination in Employment Act (“ADEA”) does not even cover job applicants. As of this writing, the case remains in discovery.
Mobley is not alone. In Baker v. CVS Health Corp., an applicant alleged that CVS’s AI video-interviewing platform violated Massachusetts law by functioning as a de facto lie detector test. That claim survived a motion to dismiss and later settled.[3] In March 2025, the American Civil Liberties Union filed a complaint against Hirevue and Intuit, alleging that an AI video-interview tool discriminated against a deaf Indigenous applicant by providing AI-generated feedback recommending that she “practice active listening.” The first AI hiring discrimination lawsuit brought by the Equal Employment Opportunity Commission (“EEOC”), against iTutorGroup, alleged that the company’s screening tool automatically rejected applicants based on age.[4] It settled.
The pattern across these cases is consistent. Plaintiffs’ firms have identified AI hiring discrimination as a viable basis on which to sue, courts have shown an increased willingness to let claims proceed past the pleading stage, and settlements are happening before verdicts. That last point matters. The cases that settle quietly do not produce the public precedent that would deter future filings. They produce the opposite: They serve as a signal to plaintiffs’ counsel that these cases have value.
The trend extends beyond employment law. On May 7, 2026, in American Council of Learned Societies v. National Endowment for the Humanities, Judge Colleen McMahon of the U.S. District Court for the Southern District of New York rejected the federal government’s argument that ChatGPT, rather than the government itself, was responsible for viewpoint-discriminatory grant terminations.[5] Judge McMahon pointedly compared the government’s defense to comedian Flip Wilson’s “the devil made me do it” catchphrase, often used to excuse his character’s wild behavior. While the ruling is a constitutional law case, not an employment law one, the rhetorical posture that employers will adopt at trial is the same and could easily meet the same fate.
The Jury Problem Nobody Is Talking About
For clients whose cases survive to trial, the liability picture looks worse than most employment lawyers have internalized. The reason is the jury pool.
AI hiring claims proceed on two tracks. Disparate-impact claims, the more common framing in cases like Mobley, turn on statistical showings and the business-necessity defense. This is territory where jury perception of AI matters less than expert testimony and the four-fifths rule. Disparate-treatment claims, by contrast, turn on whether the finder of fact believes the employer’s stated reason for the adverse action.
It is on this second track where the jury pool’s hostility to AI hiring tools creates the sharpest exposure, and where the doctrinal mechanics deserve closer attention. However, even on the disparate-impact track, juror skepticism affects how business necessity arguments rooted in AI’s claimed objectivity will land, illustrating the importance of understanding the public’s views of AI regardless of the legal theory being defended against.
Recent Pew Research Center data show that more than eight in ten American adults express concern about bias in AI-based decision-making in the hiring context, with over half describing themselves as very or extremely concerned about it. Roughly two-thirds of Americans say they would not want to apply for a job with an employer that uses AI to make hiring decisions. Most concerning for companies using AI in the hiring context is that these numbers are not driven by the demographics that typically skew concern about discrimination. They cut across partisan lines. They cut across income levels. And they cut across race and age in ways that should give defense counsel particular pause.
White non-Hispanic respondents express concern about AI bias at rates comparable to or exceeding those of minority respondents. Older Americans, those most likely to show up in a jury pool, are the most concerned of any age group, with concern among adults sixty-five and older running well above 90 percent. High-income, white-collar professionals, the jurors that defense counsel in discrimination cases typically count on to be skeptical of plaintiffs, are deeply worried about algorithmic bias in hiring.
This matters doctrinally, not just atmospherically. In disparate-treatment cases proceeding under the McDonnell Douglas burden-shifting framework,[6] once a plaintiff establishes a prima facie case, the employer must offer a legitimate, nondiscriminatory reason for the adverse action. The standard defense in AI hiring cases is some version of good-faith reliance. For example, employers can assert that they didn’t know the AI tool was biased and therefore had no reason to suspect a problem. However, under Reeves v. Sanderson Plumbing Products, Inc., a finder of fact who disbelieves the employer’s stated reason, combined with the prima facie case, may infer intentional discrimination without any additional evidence of discriminatory motive.[7]
Put the doctrinal picture together with the public opinion figures and it becomes clear why this is so dangerous for employers. A jury that overwhelmingly believes AI hiring tools are biased will be asked to evaluate the credibility of an employer’s claim that it had no reason to suspect bias in its AI systems. This illustrates the need for employers utilizing AI in the hiring process to take steps that will give them a more credible basis on which to assert nondiscriminatory intent.
Deferred Liability Is Still Liability
The Trump administration’s executive order directing federal agencies to deprioritize disparate-impact enforcement does not create a safe harbor. It simply removes one enforcement mechanism for a fixed period. Private rights of action under Title VII, the ADEA, and the Americans with Disabilities Act are unaffected. Meanwhile, multiple states, such as California, Colorado, Illinois, and Texas, are filling the enforcement gap.
What’s more, executive order priorities often reverse with changes in administrations. A new president taking office in January 2029 does not need new violations to pursue an aggressive AI hiring enforcement agenda. The next administration will inherit an actionable record of what employers are doing in the final months of the current administration.[8]
What to Do About It
The practical steps that follow can meaningfully mitigate the risk structure employers face when using AI in the hiring process.
- Audit what is deployed. Clients should know, at minimum, which AI tools are being used at which stage of the hiring process and whether those tools have ever been independently tested for disparate impact. Tools that make or heavily influence accept/reject decisions carry the highest exposure.
- Fix the vendor contracts. Indemnification clauses drafted before the agent-liability theory emerged in Mobley almost certainly do not address it. Representations about bias testing, audit rights, and data provisions should be standard negotiating points, not special asks. A vendor unwilling to provide any transparency into its testing methodology is itself a risk signal worth communicating to the client.
- Build the documentation record now. The paper trail is the defense. Clients should be maintaining records of why each tool was selected, what due diligence was conducted, what the vendor represented about bias testing, and what any subsequent review found. If an internal concern about a tool was raised and not acted on, that document will surface in discovery. The time to address it is before litigation, not during.
The Window Is Narrower Than It Looks
The current enforcement environment is not a green light for employers to proceed without caution. It is a grace period with an uncertain expiration date. Meanwhile, private litigation is proceeding regardless, the jury pool is already unfavorable, and the liability being generated today will still be there when the next administration’s EEOC decides what to prioritize. The clients who will be best positioned when the next enforcement shift comes are the ones whose lawyers helped them understand the risk now—while there is still time to do something about it.
Exec. Order No. 14281, 90 Fed. Reg. 17,537 (Apr. 23, 2025) (directing federal agencies to deprioritize enforcement based on disparate-impact liability). ↑
Mobley v. Workday, Inc., 740 F. Supp. 3d 796 (N.D. Cal. 2024) (denying motion to dismiss on agent theory); Mobley, No. 23-cv-00770-RFL, 2025 WL 1424347 (N.D. Cal. May 16, 2025) (granting preliminary collective certification under ADEA); Mobley, No. 23-cv-00770-RFL, 2026 WL 636719 (N.D. Cal. Mar. 6, 2026) (denying motion to dismiss ADEA claims). ↑
Baker v. CVS Health Corp., 717 F. Supp. 3d 188 (D. Mass. 2024). ↑
EEOC v. iTutorGroup, Inc., No. 22-cv-04128 (E.D.N.Y. 2022) (settled 2023). ↑
Am. Council of Learned Soc’ys v. Nat’l Endowment for the Humans., Nos. 25-cv-3657 (CM), 25-cv-3923 (CM), slip op. (S.D.N.Y. May 7, 2026) (consolidated). ↑
McDonnell Douglas Corp. v. Green, 411 U.S. 792 (1973). ↑
Reeves v. Sanderson Plumbing Prods., Inc., 530 U.S. 133, 147 (2000). ↑
See 42 U.S.C. § 2000e-5(e)(1) (Title VII charges must be filed within 180 days, extended to 300 in deferral states); 29 U.S.C. § 626(d) (parallel ADEA filing requirement). ↑
