Ghostwriting With AI in NJ: What RPC 8.4 Says About Deception When a Machine Drafted the Work

An attorney in a NJ small firm spends forty minutes with an AI tool, feeds it a client's fact pattern, and gets back a polished demand letter. She edits a few lines, puts her name on it, and sends it out. Nothing about that sequence is automatically unethical. But a question practitioners are not asking often enough: at what point does "AI-assisted drafting" tip into something that RPC 8.4 actually cares about?

The answer depends on context, and some of those contexts are closer than you think.

What RPC 8.4 Actually Covers

NJ RPC 8.4 defines professional misconduct. Subsection (c) prohibits conduct involving "dishonesty, fraud, deceit or misrepresentation." Subsection (d) prohibits conduct "prejudicial to the administration of justice." These aren't abstract standards, they've been applied in NJ disciplinary proceedings to situations involving misattribution and false statements to courts, clients, and opposing parties.

The rule doesn't mention AI, of course. But it doesn't need to. The underlying prohibition is on deception, and deception is deception regardless of which tool produced the underlying text.

Three Scenarios Where the Line Gets Crossed

Scenario one: the certification problem. Several NJ court forms and discovery responses require attorney certifications, the attorney affirms, under NJ Court Rule 1:4-8, that factual representations are accurate and that legal contentions are warranted by existing law or a non-frivolous argument. Signing a certification on an AI-generated document you didn't critically review isn't a billing issue or a competence issue in isolation. If the document contains a misrepresentation that the attorney could have caught with a reasonable review, and didn't, the signature becomes a false certification. That's an RPC 8.4(c) exposure.

Scenario two: representing AI output as original legal analysis to a client. If an attorney presents a legal memorandum to a client, bills for it as attorney work product, and the document is substantially unreviewed AI output containing errors or hallucinated authority, there's a colorable argument that the attorney made an implicit misrepresentation about the nature of the work. The client paid for attorney analysis; they received unchecked machine output dressed up in letterhead. NJ RPC 8.4(c) has been read broadly in disciplinary contexts, the question isn't only whether the attorney said something false, but whether the overall conduct created a false impression.

Scenario three: ghostwriting for pro se litigants. NJ has its own history with attorney ghostwriting. The use of AI adds a new wrinkle. An attorney who drafts AI-generated filings for a pro se litigant, charges for the service, and does not disclose the assistance to the court when disclosure is required runs a straightforward RPC 8.4 risk. NJ courts have not issued a definitive opinion on whether AI-assisted ghostwriting triggers additional disclosure requirements beyond what already applies to attorney-assisted pro se filings, but the court's general ghostwriting disclosure norms (discussed in the NJ Committee on Attorney Advertising's prior guidance contexts) would apply at minimum to the human attorney's involvement, and layering undisclosed AI on top compounds the opacity.

The "Mere Tool" Defense Only Goes So Far

Attorneys sometimes argue that AI is just a sophisticated word processor, that signing the output is no different than using a template or a Westlaw draft. That argument has some merit when the attorney actually reviews and takes intellectual ownership of the work. It falls apart when the attorney's review is cursory, the output is factually or legally wrong, and the attorney certifies or represents the document as sound.

The NJ Supreme Court's general guidance on attorney responsibility, reflected across the RPCs, treats the attorney as accountable for the work product regardless of how it was produced. The "tool" framing doesn't distribute responsibility to the tool.

What the rule requires

NJ RPC 8.4(c) prohibits any conduct involving dishonesty, fraud, deceit, or misrepresentation. NJ RPC 8.4(d) prohibits conduct prejudicial to the administration of justice. NJ Court Rule 1:4-8 requires that an attorney's signature on a filed document certifies the accuracy of factual contentions and the legal merit of the positions taken. These requirements apply to all work product an attorney signs or submits, regardless of how that work product was generated.

What I recommend

Before you sign anything AI drafted, read it the way you'd read a junior associate's first draft, with skepticism, not relief. Treat every factual assertion as unverified until you verify it. On court filings especially, run each cited authority independently; don't trust that the AI retrieved the right text from the right version of the rule.

If you're offering AI-assisted document services to clients, say so in the retainer or engagement letter. It doesn't have to be a lengthy disclosure, just a clear sentence that AI tools are used in drafting and that all work is reviewed and approved by the supervising attorney. That sentence does more than satisfy any emerging transparency norm, it protects you if the client later claims they were misled about what they were getting.

On the ghostwriting question: if you're drafting for a pro se party and any applicable court rule or standing order requires disclosure of attorney assistance, disclose it. The fact that AI was also involved doesn't change that obligation, and the combination of undisclosed attorney involvement plus AI-generated content is exactly the kind of compound opacity that draws disciplinary attention.

The prudent standard here isn't "can I defend this?" It's "would I be comfortable if the court, the client, and bar counsel could all see exactly what happened?" For AI-drafted work, that question is worth asking every time.