What Hallucinated Case Citations Are Actually Doing to NJ Attorneys Who File Them

A solo attorney in a New Jersey family law matter submits a motion brief. The brief cites three cases supporting the client's position on an equitable distribution issue. One of those cases does not exist. The court clerk flags it. The judge is not amused.

This scenario is no longer hypothetical. It has played out in courts across the country, and it's coming for more New Jersey practitioners as AI-assisted research becomes routine at small firms without routine verification workflows. The attorneys caught so far weren't reckless or lazy, many were competent practitioners who trusted output that looked exactly like real legal research.

That's worth sitting with for a moment, because it reframes how we should think about this problem.

The Mechanics of a Hallucinated Citation

Large language models generate text by predicting what comes next based on patterns in their training data. They don't retrieve citations from a verified database the way Westlaw or Lexis does, they produce text that statistically resembles a case citation. That means a hallucinated citation usually has all the right structural features: a plausible reporter, a plausible year, a plausible court, sometimes even a plausible judge's name. It reads like a real case.

This is different from a typo or a wrong-volume error. A typo is caught in proofreading. A hallucinated case survives proofreading because it looks correct. You have to go find the case to know it doesn't exist.

Some AI legal research tools have improved meaningfully here, platforms built on top of verified legal databases are less prone to full fabrication than a general-purpose chatbot. But "less prone" is not the same as "safe," and the reliability varies significantly by jurisdiction, practice area, and the specificity of the query. New Jersey state court decisions, particularly older ones, are underrepresented in some training datasets, which may actually make hallucinations more likely in NJ-specific research than in federal or high-volume jurisdictions.

What NJ RPC 3.3 Actually Requires

RPC 3.3(a)(1) prohibits a lawyer from making a false statement of fact or law to a tribunal. Filing a brief with a fabricated citation is a false statement of law. The intent element doesn't save you here in any clean way, if you submitted it, you implicitly represented it was real.

The more complicated question is what "knowingly" means when you used a tool that generated the citation and you failed to verify it. The NJ Supreme Court's Committee on Attorney Advertising and the Office of Attorney Ethics haven't issued AI-specific guidance on this yet as of mid-2025, but the logic of existing RPC 3.3 opinions, combined with the competence standard in RPC 1.1, points in one direction: you are responsible for what you file. The tool is not a party. The brief is yours.

Beyond ethics exposure, New Jersey courts have their own inherent authority to impose sanctions under Rule 1:4-8, which covers frivolous litigation conduct. A fabricated citation could support a sanctions motion by opposing counsel regardless of whether a formal ethics complaint follows.

What a Verification Workflow Actually Looks Like in a Solo Practice

You don't need a paralegal team. You need a fixed habit that takes roughly three to five minutes per cited case.

For every case an AI tool returns, do two things before the citation goes into a filing. First, retrieve the actual case from Westlaw, Lexis, Fastcase, or the free NJ courts website at njcourts.gov, confirm the case exists, the citation is correct, and the quoted language or holding actually appears in the opinion. Second, confirm the case is still good law using a citator (KeyCite or Shepard's). A hallucination fails at step one, so you rarely get to step two, but that's fine, step one is what catches the fabrication.

If you're using a purpose-built legal AI tool like Casetext's CoCounsel or Westlaw AI, the citations are drawn from a verified corpus and linked directly to source documents. That's meaningfully safer than prompting a general-purpose chatbot. But even with those tools, spot-checking is worth the three minutes. The liability is yours either way.

One practical adjustment: treat AI-generated research as a first draft that surfaces candidates, not as a finished cite list. Build the verification step into the drafting workflow before the brief goes to anyone else, not as an afterthought before filing. If you wait until the day before the deadline, the verification step is what gets cut.

The Disclosure Question

Some federal courts are now requiring attorneys to certify that AI-generated content has been verified. New Jersey state courts have not adopted a formal AI disclosure rule yet, but local rules and individual judge preferences are starting to matter. A few NJ judges have begun asking about AI use at case management conferences.

Getting ahead of this is straightforward. If you used AI in drafting a brief and you've verified everything, you're not harmed by saying so. If you didn't verify it thoroughly, the disclosure question is the least of your problems.

The practical next step is to write out your verification process now, as a one-page internal checklist, before you need it in a rush. That document also becomes the backbone of your firm's AI use policy, and it demonstrates, if anyone ever asks, that you took the competence obligation in RPC 1.1 seriously from the start.