New Jersey Just Made Your AI Hiring Tool a Civil Rights Liability

Photo by ThisisEngineering on Unsplash

New Jersey’s Division on Civil Rights adopted new administrative rules effective December 15, 2025 — N.J.A.C. 13:16 — implementing the New Jersey Law Against Discrimination as it applies specifically to disparate impact liability. And they made sure the rules expressly cover automated decision systems.

This is not a new law. Disparate impact theory has been federal civil rights law since Griggs v. Duke Power Co. in 1971. What New Jersey did is codify how the theory applies to AI and algorithmic systems in regulations with enforcement teeth. That matters more than it sounds.

The Legal Baseline: What Disparate Impact Means

Disparate impact liability doesn’t require proof of discriminatory intent. It requires proof that a facially neutral policy or practice produces statistically significant differential outcomes by a protected class — race, sex, religion, national origin, disability, and so on.

Under NJLAD, those protected categories apply to employment, housing, and places of public accommodation. An employer who uses a neutral screening criterion that disproportionately screens out Black applicants can be liable under NJLAD even if no one at the company wanted to discriminate against Black applicants.

The federal Title VII framework, the one that’s been litigated for fifty years, is the upstream doctrine here. NJLAD is generally interpreted consistently with Title VII but often provides broader protections.

The employer’s defense is to show the practice is job-related and consistent with business necessity. The plaintiff can then still win by showing there’s a less discriminatory alternative that would serve the same purpose.

That’s the legal framework. Now apply it to AI.

What the New Rules Add

N.J.A.C. 13:16 expressly addresses the application of disparate impact doctrine to “automated systems” and “automated decision-making systems.”

The rules make clear that when an employer uses an automated system — a resume screening algorithm, an interview scoring tool, a predictive hiring model — the disparate impact analysis applies to the outputs of that system. If the system disproportionately rejects women, or disproportionately scores Black applicants lower, that’s a cognizable disparate impact claim.

The employer can’t escape liability by pointing to the vendor. If you deployed the tool, you’re the covered entity. The fact that you didn’t write the algorithm is not a defense.

Practical Implications for Employers

If you’re a New Jersey employer using AI in hiring, you need to think about this in three stages:

1. Audit your tools. Do you know what the disparate impact analysis looks like for every AI-assisted screening or scoring tool you use? If your ATS vendor runs a resume scoring model, do you have the documentation to understand whether it produces differential outcomes by protected class? If not, you’re flying blind.

2. Talk to your vendors. The documentation you need exists — or should exist — at the vendor level. Any credible HR AI vendor should be running bias testing and providing customers with the results. If your vendor isn’t doing this or won’t share the data, that’s a problem.

3. Know what your defense would be. If challenged, you need to be able to show that the criteria your tool applies are job-related and consistent with business necessity. That means having a documented validation study — evidence that the tool actually predicts job performance and isn’t just a proxy for protected characteristics. This isn’t just good practice; it’s the legal defense.

The Difference from NYC Local Law 144

NYC Local Law 144 — which took effect in 2023 — requires employers using “automated employment decision tools” in NYC to conduct and disclose annual bias audits. It created affirmative disclosure and audit obligations.

New Jersey’s approach is different. N.J.A.C. 13:16 doesn’t require annual bias audits or public disclosure. It creates liability when you can’t defend against a disparate impact claim. The compliance incentive is to run the audit proactively so you have a defense, not because the law requires you to publish the results.

In practice, you should be doing essentially the same thing — bias testing your tools, documenting the results, validating job relatedness. But the legal structure is reactive enforcement rather than proactive disclosure.

The plaintiff’s bar will notice this. Disparate impact claims against AI systems are expensive to litigate and fact-intensive. The new rules make it clear that NJLAD applies and that the DCR will enforce it. Expect the first enforcement actions to start building a litigation record over the next few years.

If You’re a Housing Provider or Run a Business Open to the Public

The rules aren’t just for employers. NJLAD applies to housing and places of public accommodation. If you use AI tools to screen tenant applications or make lending decisions or evaluate customers in any covered public accommodation context, the same disparate impact framework applies.

A tenant screening tool that disproportionately rejects qualified applicants from protected classes is a fair housing violation, regardless of whether the landlord intended any discrimination. The tool’s outputs are what matter.

New Jersey built the legal framework to hold AI-assisted discrimination accountable under existing civil rights law. That’s the right move. The question is whether enforcement resources match the legislative intent — historically, administrative enforcement of disparate impact claims has been slow. These rules give the DCR clearer authority. Whether they use it aggressively enough is an implementation question, not a legal one.

You can find the original text of N.J.A.C. 13:16 on the New Jersey OAG’s website.