When AI Writes the Brief and a Judge Notices: How NJ Attorneys Should Handle RPC 3.3 Candor Right Now

A New Jersey attorney submits a well-organized brief. The research looks solid, the citations are formatted correctly, and the arguments flow. The judge's clerk pulls one of the cited cases and gets a 404. The case doesn't exist.

This scenario is no longer hypothetical in the broader legal world, and it's closer to home than many NJ practitioners want to admit. Courts in other jurisdictions have already sanctioned attorneys for submitting AI-generated citations that turned out to be fabricated. What hasn't been fully worked out yet, in New Jersey or anywhere else, is exactly what RPC 3.3 requires of an attorney who uses AI as part of their writing process and something goes wrong.

That ambiguity is not a safe harbor. It's a gap you need to close yourself.

What RPC 3.3 Actually Covers

New Jersey's Rule of Professional Conduct 3.3 prohibits attorneys from making false statements of law or fact to a tribunal, and from failing to correct a false statement of material fact or law previously made to the tribunal. It also requires disclosure of directly adverse controlling authority the other side hasn't cited.

The rule doesn't mention AI. It doesn't need to. The standard is output-agnostic: if a false statement of law appears in your filing, it's your statement. The fact that a language model drafted it doesn't shift the responsibility to the vendor, the software, or anyone else.

This is where a lot of small-firm attorneys are quietly miscalculating risk. They've heard about hallucinations, they've started verifying citations more carefully, and they think that's enough. It's a start, but verification workflow and disclosure obligation are two separate questions, and the second one rarely comes up in firm conversations.

The Disclosure Question Courts Are Starting to Ask

Several federal courts have now adopted standing orders requiring attorneys to disclose when AI tools were used to draft or research court filings. New Jersey's courts haven't issued a uniform AI disclosure rule as of this writing, but individual judges can and do ask, and the NJ Supreme Court's Advisory Committee on Professional Ethics has signaled ongoing attention to AI-related issues.

That environment creates a practical problem: if you don't have a default policy on AI disclosure in court filings, you're making a fresh judgment call every time, under time pressure, and possibly inconsistently across matters.

The smarter move is to decide your policy now, in writing, when you're not three hours from a filing deadline.

A reasonable starting position for NJ solo and small-firm attorneys: disclose AI use in any substantive court submission where the AI tool contributed to legal research, argument framing, or citation selection, not just final drafting. Keep the disclosure brief and factual. Something like "Portions of this brief were drafted with the assistance of AI writing tools. All citations and legal authorities were independently verified by counsel of record" is not an admission of incompetence. It's a demonstration of it.

Where the Actual Exposure Lives

The hallucinated citation problem gets the press, but it's not the only RPC 3.3 risk in AI-assisted work. Consider these scenarios that get less attention:

A general-purpose AI tool summarizes a New Jersey statute that was amended last session. The summary reflects the prior version. You don't catch it because the language sounds plausible and you're moving fast. That's a false statement of law, even if the AI generated it confidently and the error was subtle.

An AI-assisted brief characterizes a case as "holding" something the court actually noted in dicta. Depending on context and how central that point is to your argument, that characterization could become a candor issue.

An AI tool trained on national data misrepresents how New Jersey courts have applied a specific standard, because NJ case law in that area is thin and the model defaulted to majority-rule cases from other jurisdictions. Again, your name is on the filing.

None of these are exotic hypotheticals. They are the ordinary failure modes of current AI legal research tools, and they map directly onto RPC 3.3 exposure.

Building the Workflow That Protects You

The verification step most attorneys take, checking that a cited case exists and says what you claim, is necessary but not sufficient. Here's a more complete internal checklist before any AI-assisted brief goes out the door:

• Confirm every citation resolves to a real, accessible opinion (Westlaw, Lexis, or the NJ courts' own database, not just a Google result).
• Pull the full text of any case you're relying on for a specific holding and read the relevant passage, not just the headnote.
• Run any cited New Jersey statutes against the current legislative text. NJSA.gov is authoritative; AI summaries are not.
• Check whether any cited authority has been overruled, distinguished, or limited by subsequent NJ decisions, regardless of what the AI summary says.
• If the AI drafted an argument around a legal standard, confirm that standard applies in New Jersey courts specifically, not just generally.

This adds time. For a complex motion, maybe 45 minutes to an hour of focused review. That's the real cost of using AI in litigation work, and it should factor into how you staff and price AI-assisted matters.

One practical note on tool selection: AI tools built specifically for legal research, trained on verified legal data and updated regularly, carry meaningfully lower hallucination risk than general-purpose models when it comes to case law. That doesn't mean they're perfect, but the gap is real enough to matter when RPC 3.3 is the standard you're being held to.

If your current practice is using a general-purpose chatbot for case research and then spot-checking a few citations, that workflow probably isn't tight enough for tribunal submissions. Tighten it before a judge points it out for you.