Candor to the Tribunal in the Age of AI — When a NJ Litigator's Chatbot Becomes a Court Problem

By now, most New Jersey litigators have heard the cautionary tale out of the Southern District of New York — the one where a partner filed a brief citing six cases that simply did not exist, generated wholesale by ChatGPT. That story got enormous press coverage. What got far less attention is the steady drip of similar incidents since then, including filings in New Jersey federal and state courts where AI-generated citations turned out to be subtly wrong: wrong reporter, wrong holding, wrong party name, real case — wrong outcome.

The difference between "fake citation" and "real case, wrong holding" matters enormously for practicing New Jersey litigators. The first is easy to spot if you check at all. The second can slip past a tired attorney doing a quick Westlaw verify at 11 p.m. before filing.

Both, however, can put you squarely in the crosshairs of NJ RPC 3.3 — the rule on candor toward the tribunal — and its reach is longer than many small-firm litigators appreciate.

What RPC 3.3 Actually Requires (And What AI Changes About It)

RPC 3.3(a)(1) prohibits a lawyer from making a false statement of fact or law to a tribunal. RPC 3.3(a)(3) goes further: it prohibits offering evidence the lawyer knows to be false. Critically, New Jersey's version of the rule carries an ongoing duty — if you discover a misrepresentation after filing, you're obligated to correct it.

What AI changes is the mechanism by which false statements of law enter a brief. Historically, a false citation was almost always the product of a tired associate misreading a headnote or a copy-paste error. Today, a false citation can be produced in seconds by a tool that writes with complete grammatical confidence and zero epistemic humility. The AI does not know it is wrong. It does not flag uncertainty. It often produces a citation that looks indistinguishable from a real one.

That means the competence obligation embedded in RPC 3.3 — and reinforced by RPC 1.1's duty of competence — now includes understanding how your drafting tool can fail, not just whether your argument is legally correct.

What New Jersey Courts Are Actually Seeing

New Jersey federal practitioners should note that the U.S. District Court for the District of New Jersey has not yet issued a standing order specifically governing AI use the way some other federal districts have (the Northern District of Texas and the Eastern District of Michigan, for example, require AI disclosure affidavits). But that gap provides zero safe harbor. Judges in this district retain full authority to impose sanctions under Fed. R. Civ. P. 11 and 28 U.S.C. § 1927, and at least two reported colloquies in PACER-available transcripts from 2024 reflect judges asking pointed questions about whether AI was used to generate specific cited passages.

In New Jersey Superior Court, the picture is more uneven — but uneven does not mean safe. Individual assignment judges have wide latitude to manage their courtrooms, and a single adverse ruling on a motion where a citation cannot be located is, at minimum, a credibility event you cannot afford.

The Verification Gap That's Getting Attorneys in Trouble

Here is the operational problem most New Jersey solo and small-firm litigators face: AI drafting tools are genuinely useful for producing first drafts of briefs, argument sections, and motions. The time savings are real. But most attorneys using these tools have not formalized what "verification" actually means in their workflow.

Informally checking a citation on Westlaw is not verification. Verification, in the RPC 3.3 context, means:

• Confirming the citation exists exactly as stated (volume, reporter, page number, court, year)
• Reading the cited passage in the original decision — not the headnote, not a summary, the actual text
• Confirming the proposition the AI attributed to the case is what the court actually held
• Checking subsequent history — a case can exist, say what the AI claims, and still have been reversed or distinguished into irrelevance

For a 15-page brief with 30 citations, that is a meaningful time commitment. It is also non-delegable professional responsibility.

A Practical Workflow for NJ Litigators Using AI Drafting Tools

The good news: you can design this into your process once and repeat it every filing. Here is a structure that works for solo and two- to five-attorney NJ litigation practices:

1. Separate the drafting pass from the citation pass. When using an AI tool to draft argument sections, instruct it explicitly to flag every citation with a bracketed note — e.g., [VERIFY: Doe v. Roe, 342 N.J. Super. 100]. Do not treat any AI-generated citation as verified until it has been touched by a human.

2. Assign citation verification to a named person in your firm. Even if that person is you, putting it in your workflow as a discrete task — rather than assuming it happens during proofreading — reduces the chance it gets skipped under deadline pressure.

3. Use a verification log. A simple spreadsheet listing each citation, the verifying attorney or paralegal, the date checked, and the Westlaw/Lexis cite-check result creates a contemporaneous record. If a court ever questions a citation, that log demonstrates reasonable process.

4. Do not use AI to verify AI. Some attorneys have started using one AI tool to check the output of another. This is unreliable. Verification requires a primary source database — Westlaw, Lexis, Fastcase. There is no shortcut.

5. Consider a disclosure policy. The New Jersey Supreme Court's Committee on the Use of AI in the Courts is actively studying disclosure frameworks. Voluntarily disclosing AI use in your certification or cover letter, where appropriate, may actually reduce your exposure — it signals transparency and puts the verification burden on record.

The RPC 3.3 Problem Is Really a Process Problem

New Jersey attorneys who have had the closest calls with AI-generated errors share a common pattern: they were not cavalier — they were rushed. The solution is not to avoid AI drafting tools. The productivity gains are real, and in a solo practice, they can be the difference between a sustainable caseload and burnout.

The solution is to treat AI output the way you would treat a first-year associate's draft: useful, potentially impressive, but not something you file without reading every word. Build that assumption into your process now, before a judge builds it into a sanctions order for you.

---

Adam Elias is the founder of Elias Advisory LLC, helping solo lawyers and small law firms in New Jersey and across the country adopt AI tools safely, ethically, and operationally. Questions about AI governance for your practice? Get in touch.