AI Hiring in New Jersey: What Actually Binds an Employer Today

Search results will tell you New Jersey has an AI hiring disclosure law. At the time of writing it does not. The bills have been introduced and reintroduced and none has been enacted, and an employer who builds a compliance programme around a statute that does not exist will have built it around the wrong obligations entirely.

The bill is a bill

Two strands of legislation have been circulating in Trenton since the 2022–2023 session. One would require an employer to notify a candidate before an AI-analysed video interview, on the Illinois model. The other would require a bias audit of an automated employment decision tool before use, on the New York City model, with penalties per violation. Both have been reintroduced in subsequent sessions. Neither has been signed.

That matters more than it sounds, because the two models impose genuinely different work. An Illinois-style notice duty is a communications change. A New York City-style audit duty is an annual engagement with an independent auditor, a published summary of results, and a data-availability problem you have to solve before the auditor can start. Preparing for the wrong one is expensive.

What the Law Against Discrimination already covers

The binding instrument is the New Jersey Law Against Discrimination, N.J.S.A. 10:5-1 et seq. It is one of the broadest anti-discrimination statutes in the country: it prohibits discrimination in employment on the basis of race, creed, colour, national origin, nationality, ancestry, age, sex, pregnancy, familial status, marital status, domestic partnership status, affectional or sexual orientation, gender identity or expression, atypical hereditary cellular or blood trait, liability for military service, and mental or physical disability including AIDS and HIV status. Several of those categories — notably atypical hereditary cellular or blood trait and domestic partnership status — have no federal analogue, so a model screened only against Title VII categories is screened against a smaller list than New Jersey uses.

Nothing in the LAD mentions algorithms, and nothing needs to. The statute regulates outcomes and practices in hiring, and a practice does not become lawful because it was executed by software. The text is published by the New Jersey Office of Legislative Services at the Legislature’s statutes database.

The DCR guidance and its three theories

In January 2025 the New Jersey Division on Civil Rights — the agency inside the Attorney General’s office that enforces the LAD — issued guidance on algorithmic discrimination under the LAD. It does not create new obligations, and it says so; it states the Division’s position on how the existing statute applies. Three points in it are the ones that change how an employer should think about a hiring model.

• Intent is not required. The LAD reaches disparate impact as well as disparate treatment. A tool that produces a selection rate disparity can violate the statute even though nobody intended it and nobody can point to a protected characteristic in the feature set.
• Proxies are treated as the characteristic. A model that never sees race but leans on a feature that stands in for it — a postcode, a school, a gap in employment history, a commuting distance — is analysed on the effect the feature produces. The Division’s position is that using a facially neutral proxy is not a defence.
• Design choices can be disparate treatment. Where a tool is built or configured to sort on something the LAD protects, the theory is not impact at all; it is straightforward intentional discrimination executed through a model.

The guidance also addresses accommodation. A hiring assessment that depends on facial expression, voice characteristics or reaction time can disadvantage a candidate with a disability, and the LAD’s reasonable-accommodation duty applies to the selection process. In practice that means an alternative route through the assessment has to exist and candidates have to be able to find out that it does — which is a disclosure obligation arriving through the back door, even without a disclosure statute.

The vendor defence that does not exist

The single most useful line to take from the guidance is that responsibility sits with the employer. The LAD reaches an employer that uses a discriminatory tool regardless of whether the employer built it, bought it, or understands it. “The vendor told us it was audited” is a fact about your procurement, not a defence to a charge.

What that implies for contracts is concrete. The employer needs the selection-rate data to defend itself, and the vendor is usually the only party that holds it. If the contract does not give you the right to obtain scoring distributions by group, to run or commission a validity study, and to be told when the model is retrained or replaced, then you have accepted the liability without the evidence. The general shape of those terms is covered in the AI contract clauses page, and the evidence side in AI compliance evidence.

How this differs from New York City next door

New York City Local Law 144 of 2021 has been enforced since 5 July 2023, and it works on a completely different principle. It requires an employer or employment agency using an automated employment decision tool for a job or promotion in the city to have a bias audit conducted by an independent auditor within the preceding year, publish a summary of the results, and give candidates at least ten business days’ notice before use. It does not prohibit discrimination — the city’s human rights law does that separately. It is a transparency-and-process statute, and an employer can comply with it completely while still using a tool that produces a disparity. Enforcement is by the Department of Consumer and Worker Protection, which publishes its guidance at nyc.gov.

New Jersey today is the inverse: no audit duty, no notice duty, no publication duty, and full substantive liability for the outcome. An employer operating on both sides of the Hudson has to satisfy the city’s procedural regime and New Jersey’s substantive one at the same time, and the city’s audit — which measures impact ratios on the categories Local Law 144 names — is not sized to answer the LAD’s wider category list. Neither is the federal position: the EEOC’s Title VII guidance covers a narrower set of protected characteristics again.

If the New Jersey disclosure bill is eventually enacted, the effect will be to add a procedural layer on top of liability that already exists, not to create the liability. Employers who wait for the statute before doing anything are waiting for the smaller of the two obligations. The enacted-versus-pending snapshot tracks where the bill stands.