Candor to the Tribunal When AI Wrote the Brief: What NJ Attorneys Owe the Court Under RPC 3.3

A federal judge in New Jersey sanctioned an attorney last year for citing cases that didn't exist. The attorney had used an AI tool to assist with legal research and, by his own admission, didn't verify the citations before filing. The brief looked polished. The cases cited looked real. They weren't.

That incident isn't unique anymore. It's becoming a pattern, and the New Jersey Rules of Professional Conduct have something to say about it.

What RPC 3.3 Actually Requires

Most attorneys know RPC 3.3 as the "don't lie to the judge" rule. But the actual text is more demanding than that summary suggests. Under RPC 3.3(a)(1), a lawyer shall not knowingly make a false statement of law or fact to a tribunal. Under RPC 3.3(a)(3), a lawyer must correct a false statement of material fact or law previously made to the tribunal if the lawyer later discovers the error.

Notice what's missing from those provisions: intent to deceive isn't required for the problem to arise. If you file a brief with a hallucinated citation, you've made a false statement of law to the tribunal. The fact that an AI generated it doesn't create a shield. You signed the filing. You're responsible for what it says.

This is where a lot of NJ attorneys are miscalibrating the risk. They understand, in the abstract, that AI can hallucinate. What they haven't fully internalized is that a hallucinated citation in a court filing isn't just an embarrassing mistake. It's a potential RPC 3.3 violation, and depending on the circumstances, it could also implicate RPC 8.4(c) (conduct involving dishonesty or misrepresentation) if the court finds the attorney should have known better.

The Verification Problem Is Structural, Not Accidental

Here's the frustrating part: the same features that make AI useful for drafting briefs also make it dangerous for citation work. Large language models generate confident, grammatically perfect prose. They format citations in Bluebook style. They drop in quotations with parenthetical descriptions. Everything looks exactly like what a competent brief should look like.

That presentation quality creates a verification problem that's actually harder to manage than a first draft from a junior associate. When an associate cites a case, attorneys instinctively check it, because they know the associate might have made a mistake. When the AI produces a citation that's formatted perfectly and sounds authoritative, the instinct to verify gets weaker. The output looks done.

The fix isn't complicated, but it does require a deliberate workflow change. Every single case citation generated by an AI tool needs to be verified in Westlaw, Lexis, or Fastcase before the document gets filed. Not spot-checked. Every one. The verification step needs to confirm three things: the case exists, the quoted language is accurate, and the holding is actually being characterized correctly in context.

That last point matters more than attorneys often realize. AI tools will sometimes cite a real case but describe its holding in a way that's subtly wrong, or cite a case that was subsequently reversed without flagging that fact. Checking that the citation exists isn't enough.

What NJ Courts Are Watching For

The NJ District Court adopted local rules in 2024 requiring disclosure of AI use in certain filings. Some NJ judges have gone further in individual standing orders. This is a developing area, and the procedural requirements vary by court and judge.

But beyond the disclosure logistics, what courts are actually reacting to is a pattern of attorneys filing documents without adequate quality control. The sanction risk isn't hypothetical anymore. In addition to the NJ federal case mentioned above, similar sanctions have come down in the Southern District of New York, the Eleventh Circuit, and Texas federal courts. The judicial tolerance for "I didn't know AI could do that" explanations has shortened considerably.

A Practical Verification Protocol for NJ Firms

If you're using AI to assist with brief writing, build the following into your process before anything gets filed:

Run every citation through a primary legal research database. Not a secondary summary. The actual case.

Read the specific language you're quoting in the original opinion. AI tools will sometimes paraphrase rather than quote, and the paraphrase occasionally shifts the meaning.

Shepardize or KeyCite every case cited for current validity. An AI has a training cutoff and no awareness of subsequent history.

If opposing counsel or the court flags a citation problem after filing, address it immediately under RPC 3.3(a)(3). Waiting to see if anyone notices is not a strategy.

Some firms are now assigning a non-AI-assisted attorney review of all citations before filing, treating it the same way they'd treat a final proofreading pass. That's a reasonable standard given the current stakes.

The Competence Angle Connects Here Too

RPC 1.1 requires competent representation, which includes keeping up with changes in the law relevant to practice. The NJ Supreme Court's Advisory Committee on Professional Ethics hasn't yet issued a formal opinion specifically on AI-generated filings, but the ACPE's guidance on competence has consistently emphasized that attorneys are responsible for the work product they submit regardless of who or what produced the first draft.

Using AI to draft briefs isn't the problem. Filing AI output without adequate review is. The distinction matters ethically, and it matters practically because it's exactly the distinction courts will draw when deciding whether sanctions are appropriate.

If you're using AI tools in your NJ practice for litigation work, the one concrete step worth taking right now is writing down your citation verification procedure and making sure everyone in the firm who touches court filings follows it. That document is also your best evidence of a reasonable quality-control process if a question ever arises.