Candor to the Tribunal Gets Complicated: What the NJ Supreme Court's AI Guidelines Actually Say About Court Filings

Ask a solo attorney what scares them most about using AI for court filings and you'll hear some version of the same answer: "Getting caught submitting something the AI made up." That fear is reasonable. But the actual framework governing what NJ attorneys owe the court when AI is involved is more specific than most practitioners seem to realize, and ignoring the specifics is where the exposure lives.

There are two layers to understand. First, the longstanding rule. Then, the NJ Supreme Court's more recent guidance that sits on top of it.

The RPC 3.3 baseline

NJ RPC 3.3(a)(1) prohibits a lawyer from knowingly making a false statement of fact or law to a tribunal. RPC 3.3(a)(3) goes further: a lawyer may not offer evidence the lawyer knows to be false. These aren't new principles. What's new is that AI tools can now generate plausible-sounding case citations, statutory quotations, and factual characterizations that are simply wrong, and they can do it with no visible signal that anything is amiss.

The rule's trigger word is "knowingly." A hallucinated citation a lawyer submits without checking is a harder case than a deliberate lie, but it's not a clean escape. Courts in other jurisdictions have found that filing AI-generated content without verification crosses into recklessness, and recklessness can satisfy the "knowingly" standard when a reasonable attorney would have discovered the error. NJ RPC 3.3 does not carve out an AI exception.

What the Preliminary Guidelines add

On January 24, 2024, the NJ Supreme Court issued its "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence." The document isn't a set of standalone rules, but it does map existing RPC obligations onto AI-specific conduct. On court submissions, the guidelines are direct: attorneys are responsible for the accuracy of AI-generated content in any document filed with a court, and that responsibility requires verification before filing.

The guidelines also flag what they call "confidently wrong" output, the tendency of generative AI to produce authoritative-sounding text that contains factual errors. They place the burden of detecting those errors squarely on the submitting attorney, not on the tool, not on the vendor, and not on a paralegal who ran the prompt.

That word "verification" carries weight. A quick re-read of a brief is not the same as independently confirming that every case cited actually exists, says what the brief claims it says, and remains good law. The guidelines don't specify a verification protocol, which means attorneys have to construct their own. The standard against which that protocol will be measured is the competent practitioner standard under RPC 1.1, a subject this blog has covered before, but the application here is distinct: the question is whether your pre-filing review was sufficient to catch the specific failure modes of the AI tool you used.

The certification problem most NJ attorneys are skipping

NJ Court Rule 1:4-8 governs attorney certifications on court filings. By signing a document, an attorney certifies that the claims and legal contentions are warranted by existing law or a non-frivolous argument for modification of existing law, and that factual contentions have evidentiary support. A filing built on an AI-hallucinated case citation fails that certification on its face.

The Preliminary Guidelines don't directly amend Rule 1:4-8, but they make plain that AI-assisted drafting doesn't change what the certification covers. If you sign it, you own it. This is worth reading carefully for attorneys who hand AI-drafted briefs to a partner or associate for a signature: the signer is the one exposed, not the attorney who prompted the tool.

A practical hypothetical

Say a solo family law attorney uses an AI drafting tool to prepare a motion brief. The tool generates a case cite from a different jurisdiction that doesn't apply in NJ and never did. The attorney reviews the brief for argument flow but doesn't run each citation through Westlaw or Fastcase. The brief gets filed. Opposing counsel flags the error at oral argument.

Under RPC 3.3(a)(1) and the Preliminary Guidelines' verification requirement, that attorney has a problem. The fact that AI produced the error doesn't shift the duty. The certification under Rule 1:4-8 doesn't have an AI carve-out. And if the error caused any prejudice to the opposing party or consumed court resources, the court has tools available, including sanctions under Rule 1:4-8(d), that don't require a formal disciplinary referral.

What the rule requires

NJ RPC 3.3(a)(1) and (a)(3) prohibit knowingly false statements of fact or law and the knowing submission of false evidence to any tribunal. NJ Court Rule 1:4-8 requires attorneys to certify, by signature, that legal contentions are warranted by law and factual assertions have evidentiary support. The NJ Supreme Court's Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence (January 24, 2024) establish that attorneys bear full responsibility for verifying the accuracy of AI-generated content before submitting it to a court, and they identify "confidently wrong" AI output as a specific risk attorneys must address.

What I recommend

Before any AI-assisted brief or motion leaves your desk, build a hard stop into your workflow: every case citation gets verified independently, not by asking the AI to confirm its own output, but through a legal research database that retrieves the actual decision. Every statutory quotation gets checked against the official text. That step should be logged somewhere, even a line in a file note, so that if a question arises later you can show the verification happened.

If you use a paralegal or a virtual assistant to run AI drafts, be explicit in writing that citation verification is their responsibility before anything reaches you, and then you verify again. The Preliminary Guidelines' supervision language means the duty runs up to you regardless.

Finally, consider building a pre-filing checklist specific to AI-generated documents. It doesn't have to be long: Does every cited case appear in Westlaw or Fastcase? Does each quote match the source? Does each cited case remain good law? Has the AI been asked to generate anything it might have invented rather than retrieved? That last question sounds odd, but asking an AI "did you retrieve this citation or generate it from training data?" can surface uncertainty the tool would otherwise paper over. It's not foolproof, but it's one more filter before your signature goes on the page.