New Jersey tenant screening AI: the rules already in force

If you run tenant screening in New Jersey, or you build the software that does it, the question worth asking is not whether a new AI bill will pass. It is whether your current screening process would survive a disparate impact claim under rules that took effect in December 2025.

That distinction matters because of how the news usually arrives. A bill gets introduced, a round of coverage describes what it would require, and the compliance conversation shifts to a future deadline. Meanwhile the state civil rights agency has already told you, in writing, that your screening algorithm can violate state law today, with no new statute needed.

This post covers both: what New Jersey's Assembly Bill A5495 is, what remains unknown about it, and the law already in force that most screening operations have not adjusted for.

What A5495 is, and what is not yet public about it

Assembly Bill A5495 was introduced on 14 September 2026 in the 222nd Legislature. Its sponsor is Assemblyman Al Abdelaziz, and as of this writing he is the only sponsor listed. It went to the Assembly Science, Innovation and Technology Committee, where it remains pending. No committee vote has been recorded.

The legislature's own synopsis is one line: it regulates the use of automated tenant screening tools in evaluating prospective tenants.

That is the extent of what is confirmed. The introduced text is not currently retrievable from the legislature's public site, so the bill's operative provisions are not yet a matter of public record, and a one-line synopsis will not tell you what duties it would create.

This matters because of where New Jersey's bias audit language sits. The requirement people associate with these proposals, phrased as a prohibition on selling or using a tool not audited within the preceding year, belongs to the state's employment bills. New Jersey has run several attempts at regulating automated employment decision tools on that model, borrowed from New York City. None of them governs tenant screening, so check any description of A5495 as an annual independent audit mandate against the bill itself once the text is public.

One citation trap. Bill numbers recycle between sessions, and a different A5495 from the 2022 to 2023 session was enacted as P.L.2023 c.330, a law about which firearms may be carried while hunting. Nobody will confuse the two on subject matter, but a search for "A5495" surfaces both, so check the session before citing it.

So the honest status of A5495 is: real, early, narrow in sponsorship, procedurally almost nowhere. One sponsor, no committee vote, no public text. The useful reading is as a signal of legislative attention, not as a compliance deadline.

Tenant screening algorithms are already exposed under state law

New Jersey's Law Against Discrimination already reaches algorithmic tenant screening, and the state has said so explicitly.

In January 2025, the Attorney General and the Division on Civil Rights issued guidance on algorithmic discrimination. It covers housing alongside employment, credit, public accommodations and contracting. Three points in it have direct consequences for screening:

• Intent is irrelevant. A covered entity can violate the LAD through a tool that produces discriminatory outcomes, whatever the motive behind deploying it.
• Using a vendor's tool is not a shield. The guidance is direct that a third party's role in building or operating the tool does not transfer your liability away.
• Uneven scoring is not the only route to liability. A tool that obstructs a required reasonable accommodation creates exposure of its own.

The guidance is not speaking about housing in the abstract. Two of its worked examples are tenant screening: a tool designed to exclude prospective tenants who use housing assistance, and running a screening algorithm only on Black applicants while other applicants are assessed another way.

The guidance also states it imposes no obligations beyond what the statute already contains. People misread that as reassurance. It means the exposure was always there.

Then in December 2025 the Division adopted disparate impact rules, effective on publication in the New Jersey Register on 15 December 2025. Under them, a facially neutral practice that disproportionately harms a protected class can be unlawful unless it is necessary to achieve a substantial, legitimate interest and no less discriminatory alternative is available. Note what the rules do not say: an alternative need not be identical in cost or efficiency to count, a drafting choice the Division made deliberately.

There is an affirmative duty attached. Where the practice relies on an outside vendor's products or systems, you must take reasonable steps to ensure those are consistent with the statute in their own right. Diligence on your vendor is part of the obligation, not a precaution.

The rules are in effect, though not unchallenged. A mortgage industry trade group has sued in federal court in New Jersey, arguing the rules are preempted by the Fair Housing Act and the Equal Credit Opportunity Act. That case is unresolved, and the rules bind while it runs.

The rules include a housing example that reads like it was written for tenant screening, because it was. The practice they name is automatically refusing every rental applicant who has no credit score, or a credit score below a minimum threshold. The alternative they point to is an individualised assessment of the applicant's credit history alongside other information bearing on their ability to reliably pay rent.

Read that example next to how most screening stacks are configured. A minimum-score cutoff, applied uniformly, with no individualised review path, is precisely the pattern the rules single out. The question a New Jersey landlord should be asking is not what A5495 might require next year. It is whether that cutoff can be defended as necessary when the state has already named the alternative.

What New York City's bias audit experiment produced

One jurisdiction has a track record here. New York City's Local Law 144 has required bias audits for automated employment decision tools since 2023, and the compliance checklist for employers sets out what that involves in practice.

Anyone expecting an audit mandate to deliver transparency on its own should read the results.

A New York State Comptroller audit released on 2 December 2025 found the Department of Consumer and Worker Protection's complaint process for the law ineffective, and its wider enforcement approach unable to identify non-compliance. The specifics are stark. DCWP reviewed 32 companies and identified a single instance of non-compliance. The Comptroller's own team, examining those same companies, identified at least 17 instances of potential non-compliance. Over the two years the audit covered, DCWP received two AEDT complaints in total.

Independent research found a similar gap on the employer side. The "Null Compliance" study, presented at ACM FAccT in 2024, had student investigators check 391 employers drawn from companies that recruit at Cornell for evidence of LL144 compliance. Eighteen had posted an audit report. Thirteen had posted the required transparency notice. The authors are careful that a missing report is not proof of non-compliance. They attribute the rates to limitations in the law's accountability mechanisms, not to employer indifference alone.

The published audits are also thin on the ground. "Auditing the Audits," presented at ACM FAccT in 2025, assembled every public bias audit report it could find as of early November 2024 and reached 44 reports covering 116 audits.

Bias audits can still do useful work. What this record shows is that an audit obligation without a functioning disclosure and enforcement mechanism produces little observable accountability. If New Jersey legislates here, the design questions are who checks, how, then what happens when nobody posts anything. Those are harder to answer than an annual cadence.

The other New Jersey bill that may matter more

While A5495 sits in committee with one sponsor, a Senate bill takes a different route to the same problem.

S4279, introduced on 14 May 2026 by Senator Troy Singleton, would create a disparate impact cause of action based on automated decision systems. Its findings name tenant screening explicitly, alongside mortgage lending, credit underwriting, rental pricing. They go on to describe how facially neutral systems produce disparate outcomes through data selection and proxy variables.

Reporting on the introduced bill describes three features that would reach further than an audit mandate: a civil penalty with a floor instead of a ceiling, a right for individuals to request meaningful human review of an adverse automated decision, plus an express carve-out so that nothing requires disclosure of proprietary source code or trade secrets. The bill text would not load for us, so treat those specifics as provisional until you read it.

Watch the third one. The standard industry argument against algorithmic accountability laws is intellectual property exposure. A bill drafted to remove that objection is harder to oppose on familiar ground.

Its status, like A5495's, is introduced with no recorded advancement. Neither bill should be planned around as a certainty. Both indicate where New Jersey's attention is.

What to do before any of this passes

The useful work here does not depend on which bill moves, because it is driven by the rules already in force. Each of these is worth doing on the December 2025 disparate impact rules alone.

Find every hard cutoff and ask whether it can be justified. Score thresholds, income multiples, automatic criminal or eviction record denials. For each, the rules force two questions: is this necessary to a substantial legitimate interest, then is there a less discriminatory alternative. The credit-score example tells you how the state answers the second question for at least one common practice. Start here, across every tool that touches a housing decision, including the screening vendors, credit scoring, income verification, fraud models and rental pricing you did not build yourself. The vendor relationship does not move the liability.

Build an individualised review path. The alternative the rules name is individualised assessment of ability to pay. A process that can only approve or deny on a score has no answer to that.

Test for disparate impact, and keep the results. Measurement you did not run is not evidence you can produce later, and the disparate impact framework puts the burden of justification on you well before any audit statute arrives.

Get your notices and records in order. S4279 ties its penalty partly to notice failures. Adverse action notice duties already exist under federal credit reporting law for most screening workflows.

Pin down what your vendor will tell you. Ask for impact testing, the protected-class breakdowns behind it, and contractual support for your own compliance. The LL144 record suggests you should not assume a report exists because a law says it should.

None of this requires a new statute to be worthwhile. All of it is what you would need anyway if a claim arrived under the rules as they stand.

Where this is heading

New Jersey is not deciding whether to regulate algorithmic housing decisions. The Division on Civil Rights settled that in January 2025 and reinforced it with the December 2025 rules. The open question is whether the legislature adds a procedural layer on top, plus what shape it takes.

The reasoning also travels. Disparate impact analysis is not a New Jersey invention, so a screening practice that cannot survive the necessity test here is exposed anywhere a regulator applies the same framework. Our survey of state AI governance regulation across the United States tracks where that is heading.

For screening operators, today's enforcement risk runs through civil rights law, not a pending bill. For proptech vendors, three things follow: New Jersey's attention has moved to tenant screening specifically, your customers now carry disparate impact exposure they may not have priced, plus S4279's trade-secret carve-out suggests legislators have learned which objections to design around.

VerifyWise tracks obligations and evidence across jurisdictions, including the disparate impact exposure these rules create. For help scoping what applies to your screening stack, get in touch.