New Jersey Takes The Lead On AI Accountability In The Workplace

Highlights:

• Employers remain legally responsible for biased AI outcomes even when using third-party hiring platforms.
• New Jersey regulators are treating AI-driven discrimination exactly like traditional workplace discrimination.
• AI systems that ignore accommodations or historical bias data may create immediate legal exposure for employers.

Employers are increasingly turning to artificial intelligence to streamline hiring and workforce management. A 2024 Rutgers University survey found that 63% of the New Jersey employers who participated now use AI-enabled tools to recruit or make hiring decisions¹. In light of its growing prevalence in the workplace, New Jersey has emerged as a frontrunner in regulating AI decision-making in employment. Through administrative guidance issued in January 2025 and new regulations finalized in December 2025, the New Jersey Attorney General and Division on Civil Rights have sent a clear message: AI tools are not a shield against liability under the state's Law Against Discrimination (“LAD”).

The 2025 Guidance on Algorithmic Discrimination – Misusing AI

In January 2025, the New Jersey Office of the Attorney General and the Division on Civil Rights (“DCR”) issued guidance to clarify how the LAD applies to “algorithmic discrimination” –discrimination resulting from the use of automated decision-making tools (the “Guidance”)². The Guidance confirmed that the LAD applies to algorithmic discrimination the same way it applies to other discriminatory conduct, prohibiting bias in employment on the basis of race, religion, national origin, sex, gender identity, disability, and other protected characteristics.

The Guidance defined automated decision-making tools broadly to mean any “technological tool, including but not limited to, a software tool, system, or process that is used to automate all or part of the human decision-making process.”³ Such automated decision-making tools are increasingly used to determine whether a human reviewer reads a resume, whether an applicant is hired, and whether an employee is promoted, demoted, or fired, among other things. The DCR explained that such tools often accomplish their aims by using algorithms that analyze data, uncover correlations, and make predictions based on those correlations⁴. By doing so, however, these tools can create classes of individuals who will be either advantaged or disadvantaged in ways that may exclude or burden them based on their protected characteristics.

The Guidance discussed three primary stages where bias can impact automated decision-making tools: (1) design, (2) training, and (3) deployment.

1. Design choices, including how the developer translates a real-world problem into something (often numerical) that can be analyzed by an automated decision-making tool, can skew the tool purposefully or inadvertently in a discriminatory manner. The outputs a tool provides, what algorithm it uses, and what inputs it assesses are all design choices that can lead to biased recommendations.⁵
2. Many AI programs must be “trained” on a data set before they can be implemented. If tools are trained on data that is skewed, unrepresentative, or reflects historical bias, it can cause the tool itself to become biased.⁶
3. The way an automated decision-making tool is deployed can contribute to algorithmic discrimination if the tool is used to make decisions it was not designed to make or is used selectively, for members of one protected class but not another.⁷

The Guidance emphasized that employers may be held liable even if they use a tool they did not develop. Put differently, an entity is not shielded from liability simply because a third party developed the tool or because the entity does not understand its inner workings.⁸

New Disparate Impact Regulations

Building on this foundation, the DCR finalized new disparate impact regulations in December 2025 (the “Regulations”)⁹. The Regulations codify existing legal rights and precedents regarding disparate impact discrimination under the LAD and provide explicit guidance on automated employment decision tools. “Disparate impact” refers to a practice or policy that is facially neutral but actually or predictably results in a disproportionately negative effect on members of a protected class, as shown by empirical evidence – regardless of intent.¹⁰

The Regulations incorporate a burden-shifting framework for disparate impact claims brought in the employment context¹¹. If a complainant demonstrates that a practice or policy has a disparate impact on members of a protected class, the burden shifts to the employer to show the practice is necessary to achieve a substantial, legitimate, nondiscriminatory interest. Even if the employer meets this burden, a complainant may still prevail by showing there is a less discriminatory alternative that would achieve the same interest¹². A “substantial, legitimate, nondiscriminatory interest” means one that is a core interest of the employer, is genuine and not pretextual, and is not itself discriminatory.¹³

The Regulations explicitly reference the use of automated decision-making tools in the employment context¹⁴. The regulations explain that the use of automated decision-making technology can cause a disparate impact on applicants and employees based on their race, national origin, gender, disability, religion, and other protected characteristics and provide examples of such algorithmic discrimination.

Notably, the Regulations also address vendor relationships. If an employer’s practice or policy that results in a disparate impact relies on conduct, standards, procedures, or systems of an outside vendor, the employer must take “reasonable steps” to ensure that the vendor’s conduct or product is consistent with the LAD¹⁵. The DCR’s comments to the Regulations also clarify that taking reasonable steps to vet a third-party’s product is an affirmative requirement to using such automated decision-making tools and is not a complete defense against LAD claims if employers do not prevent or mitigate disparate impact discrimination.¹⁶

Potential Pitfalls for Employers

Perhaps the most common mistake employers can make when incorporating AI or other automated decision-making tools into their hiring and management decisions is failing to consider reasonable accommodation requirements. When an automated decision-making tool has not been trained on data reflecting individuals who use accommodations, it may fail to account for the possibility of an accommodation or may negatively score individuals who require one. If a covered entity relies on the output of such a tool, it risks engaging in disparate impact discrimination.

In the hiring context, for instance, the tool may disproportionately screen out applicants who are fully capable of performing the essential functions of the job with a reasonable accommodation. In the context of management decisions, if an automated decision-making tool is used to monitor and track employee productivity, it may unfairly penalize certain employees for taking breaks that appear unsanctioned or atypical. This risk arises when the tool has not been programmed or trained to account for reasonable accommodations that entitle those employees to additional break time, such as breaks necessitated by a disability or for lactation purposes.

Other potential pitfalls for employers are neglecting to continuously monitor their automated decision-making tools, failing to test for bias before deployment, and assuming the vendor of such products is liable for any discriminatory outcomes.

Finally, the Regulations emphasize that employers who place excessive reliance on efficiency-based justifications may be exposed to liability for algorithmic discrimination. Although the Regulations retain the business necessity defense, the DCR clarifies that a less discriminatory alternative need not be equally effective in promoting the employer’s interest as the challenged practice. This means an employer cannot refuse to adopt a less discriminatory approach merely because it involves somewhat greater expenditures of labor, time, or resources.

Employers should consult with legal counsel in advance of implementing AI or other automated decision-making tools in their hiring and management practices to ensure they do not create unnecessary exposure under the LAD.

Footnotes

1. Liu, et al., Shaping the Future of Recruitment: A Survey on AI-enabled Hiring Tools, 4 (2024). Retrieved April 21, 2026, from https://smlr.rutgers.edu/sites/default/files/Documents/Faculty-Staff-Docs/Shaping_the_Future_of_Recruitment_Report_Rutgers.pdf.

2. New Jersey Office of the Attorney General and New Jersey Division on Civil Rights, Guidance on Algorithmic Discrimination and the New Jersey Law Against Discrimination, (January 2025). Retrieved April 21, 2026, from https://www.nj.gov/oag/newsreleases25/2025-0108_DCR-Guidance-on-Algorithmic-Discrimination.pdf.

3. Id. at 2.

4. Id. at 3.

5. Id. at 6.

6. Id. at 7.

7. Id. at 8.

8. Id. at 9.

9. New Jersey Division on Civil Rights, Rules Pertaining to Disparate Impact Discrimination, N.J.A.C. 13:16 (December 15, 2025). Retrieved April 22, 2026, from https://www.njoag.gov/wp-content/uploads/2025/12/DCR-Disparate-Impact-Discrimination-Rules-13_16-12.15.2025.pdf.

10. N.J.A.C. 13:16-2.1(b).

11. N.J.A.C. 13:16-3.1.

12. N.J.A.C. 13:16-2.2.

13. N.J.A.C. 13:16-2.4.

14. N.J.A.C. 13:16-3.2(c).

15. N.J.A.C. 13:16-2.4(e).

16. New Jersey Division on Civil Rights, Rules Pertaining to Disparate Impact Discrimination, Response to Comment 28 at 43 (December 15, 2025). Retrieved April 22, 2026, from https://www.njoag.gov/wp-content/uploads/2025/12/DCR-Disparate-Impact-Discrimination-Rules-13_16-12.15.2025.pdf.

Originally published by HR.com.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.