Unprovable Discrimination in AI Hiring Practices
Sam Peterson
As of January 2026, 99 percent of Fortune 500 companies use artificial intelligence in some capacity during their hiring processes. [1] As AI continues to develop, companies are expected to increase its usage to streamline their hiring. AI tools have helped automate certain hiring tasks, such as resumé reviewing, candidate screening, and matching applicants to specific roles. Data collected on specific candidates can be used to determine their ‘fit’ for a role and, in each company, can help match applicants with specific opportunities. Artificial intelligence is even used, in some cases, to conduct initial interviews through video calls. AI’s utility in hiring is touted for reducing time-to-hire, streamlining processes, providing financial advantages, and enabling objective hiring to find the highest-quality candidate, while avoiding potential human biases. The use of AI tools has become an essential part of the hiring process for many companies of all sizes. This paper will first show how artificial intelligence used for hiring has the potential to create bias. Then, it will demonstrate that these tools should be held accountable, and how current precedent may prevent that accountability. After examining the flaws surrounding current rulings regarding AI in hiring, this paper will propose a solution that shifts the burden to prove fairness to the employer using the AI system, thereby mitigating any potential bias caused by these systems and complying with the Civil Rights Act of 1964.
Since the initial implementation of AI hiring tools with the digitization of the hiring process, issues have stemmed specifically from their inability to avoid bias. AI models are trained on pre-existing datasets of applications. However, if the datasets themselves are imbalanced, the AI training on them will be as well. A notable example of this occurrence was Amazon’s since-discontinued hiring tool, which held a bias favoring masculine language due to primarily being trained on men’s resumés. [2] Today, artificial intelligence has become much more complex than the system that Amazon used, and different tools can be biased with less obvious results.
The Equal Employment Opportunity Commission (EEOC), established under Title VII of the Civil Rights Act of 1964, exists to ensure that hiring processes do not disproportionately negatively impact its protected groups. [3] The ruling of Griggs v. Duke Power Co. (1971) offers protection from even seemingly innocent hiring practices that lead to disproportionate hiring results with the idea of disparate impact. [4] The test to determine if the disparate impact ruling is applicable contains three parts: prima facie, business necessity, and the proposal of an alternative practice. Prima facie, or the first impression of a certain practice, requires the plaintiff to prove that a neutral practice leads to disparate outcomes based on race, religion, and gender, among other factors. [5] To do so, the plaintiff must show data that proves the practice has disproportionate adverse impacts on a protected group, as well as demonstrate how a specific practice leads to that disparate result. The burden is then placed on the defendant, the business, to prove that the practice in question is necessary for the role they are hiring for. Finally, the plaintiff must provide an alternative practice that achieves the business’s goal without being discriminatory. However, the obscurity of the specific workings of AI hiring systems to both the employer and the candidate means that the plaintiff cannot prove disparate impact according to the current test. In cases where AI hiring tools cause the alleged disparate impact, the test to determine disparate impact established by Griggs is ineffective, as the plaintiff is unable to provide the prima facie component of the test due to the obscurity of black box AI hiring systems, so called due to the opacity of their operations. Instead, the burden of proof must be placed initially on the employer to protect the EEOC’s protected groups.
First, it must be established whether Title VII of the Civil Rights Act of 1964 applies to AI hiring tools. Title VII of the Act states that employment discrimination against its protected groups is illegal. [6] Additionally, Griggs established that employment requirements that lead to disparate impacts, even unintentionally, are illegal. [7] The EEOC maintains guidelines on fair hiring practices, even when applicants are referred by an external hiring agency. [8] AI tools make hiring decisions for employers; therefore, although they are external tools, the company using the tool can be liable for any bias stemming from their outsourced hiring practice. Employers may still be liable for biased hiring from actions such as resumé screening and video interviews conducted by AI agents.
Although discriminatory outcomes might be apparent, the plaintiff may be unable to prove which specific part of the AI’s process caused it. To hold employers using biased AI agents liable, the affected applicant or employee must first establish the prima facie case. Following the guidelines established in Griggs, prima facie requires demonstrating how a specific, facially neutral employment practice causes bias towards a protected group. [9] Black box AI tools can lead to biased results due to several factors: biased training data, when the AI is trained on historically biased data; skewed representation, when the training data does not represent certain groups; the use of proxy variables, where an applicant's race, gender, or other information is inferred through their application details even if it is not included in their application; and misaligned objectives, when an AI tool might prioritize “cultural fit” in a workplace. [10] The decision-making processes of these systems are complex and not easily understood, especially when an employer outsources an AI hiring tool. Disparate outcomes, however, are more easily proven; a 2026 study from Stanford University found that around 25 percent of Black applicants, from a pool of a total of three million applicants applying through an AI hiring tool, faced discriminatory hiring results according to the EEOC’s definition. [11] Under Griggs, the plaintiff must identify a specific practice, a disparate outcome, and the causal link between the hiring practice and disproportionate result. Although the disproportionate results may be readily identified, the AI’s opacity prevents the plaintiff from demonstrating which factor causes the facially neutral AI to be biased in its hiring.
Because the plaintiff may be unable to prove the prima facie component of their disparate impact claim, even if they have experienced biased hiring results, the burden of proof that their hiring system results in fair outcomes should be placed on the employer. As it stands, Griggs, although intended to “[proscribe] not only overt discrimination, but also practices that are fair in form, but discriminatory in operation,” may fail to do so in cases where black box AI systems obscure which specific hiring practice causes discrimination. [12] An alternative method, implemented in 2021 by the city of New York, preempts AI hiring discrimination by requiring employers using automated employment decision tools (AEDTs) to commission independent bias audits, publish their results, and inform subjects of the AEDT of its function and provide them an alternative selection process. [13] As opposed to disparate impact testing to prove discrimination with Griggs, N.Y.C. 144 (2021) ensures unbiased AI hiring tools to begin. Because the plaintiff is likely unable to prove how exactly the AI hiring tool discriminated against them, N.Y.C. 144 (2021) shifts the burden to the employer to prove that their AI hiring tool is verifiably unbiased.
Artificial intelligence has emerged and is incorporated in every part of the hiring process for many companies, and will continue to be further incorporated. As the technology develops, the federal law must develop with it. Under the Civil Rights Act of 1964, the federal government must enforce equal opportunities in hiring; the EEOC’s current guidelines confirm that the agency plays a role in ensuring fair hiring with AI hiring tools. [14] In fact, these guidelines explicitly caution employers on how AI tools might lead to disparate impact, even though they are “seemingly neutral employment practices.” [15] However, the current tests required to prove disparate impact by the EEOC may fail to prevent discrimination against protected groups–black box AI systems obscure their processes, meaning that affected applicants may be unable to pinpoint the exact practice causing the discrimination. To maintain fair hiring practices, the mandates in N.Y.C. 144 (2021) must be adopted at the federal level. By doing so, employers will continue to be able to use AI tools to streamline their hiring processes while ensuring equal opportunities to protected groups. The implementation of these standards at the federal level would convert AI tools from unregulated and potentially discriminatory to unbiased hiring agents, as intended.
[1] Akashdeep Singh, “AI Recruitment Statistics (2026) – Hiring Trends & Market Size,” DataRefs, January 17, 2026, https://www.datarefs.com/statistics/ai/ai-recruitment/.
[2] Jeffrey Dastin, “Insight - Amazon Scraps Secret AI Recruiting Tool That Showed Bias against Women,” World, Reuters, October 11, 2018, https://www.reuters.com/article/world/insight-amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK0AG/.
[3] Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e (1964). https://www.law.cornell.edu/uscode/text/42/2000e
[4] Griggs v. Duke Power Co., 401 U.S. 424, 430 (1971) https://www.law.cornell.edu/supremecourt/text/401/424.
[5] Griggs, 401 U.S at 431
[6] Title VII of the Civil Rights Act of 1964, 42 U.S.C.
[7] Griggs, 401 U.S. at 430.
[8] U.S. Equal Employment Opportunity Commission, “Coverage of Employment Agencies,” Accessed May 20, 2026. https://www.eeoc.gov/employers/coverage-employment-agencies.
[9] Griggs, 401 U.S. at 431.
[10] Kathy Heldman, “Workplace Fairness, Empower Worker - Employment Law,” Workplace Fairness, Empower Workers, January 20, 2025, https://www.workplacefairness.org/.
[11] Rishi Bommasani, Sarah H. Bana, Kathleen A. Creel, Dan Jurafsky, and Percy Liang, "Algorithmic Monocultures in Hiring," arXiv, May 26, 2026, https://arxiv.org/abs/2605.27371.
[12] Griggs, 401 U.S. at 431.
[13] N.Y.C Local Law 144 § 20-871 (2021).
[14] “What is the EEOC’s role in AI?”, accessed May 29, 2026, https://www.eeoc.gov/sites/default/files/2024-04/20240429_What%20is%20the%20EEOCs%20role%20in%20AI.pdf.
[15] “What is the EEOC’s role in AI?”
