Supervising a Non-Lawyer Who Uses AI: What NJ RPC 5.3 Actually Requires

A paralegal at a two-attorney family law firm in Cherry Hill runs every first draft of a client letter through ChatGPT. She prompts it, reviews the output for about ninety seconds, and sends the document to the supervising partner for a quick sign-off. The partner glances at the formatting, sees nothing obviously wrong, and approves it.

That workflow has a name in New Jersey ethics rules: a problem waiting to surface.

NJ RPC 5.3 requires attorneys who supervise non-lawyers to make "reasonable efforts to ensure that the person's conduct is compatible with the professional obligations of the lawyer." Most NJ solo and small-firm attorneys know this rule exists. Very few have thought seriously about how it applies when the non-lawyer has an AI tool in the loop.

The Rule Predates AI, But It Adapts

RPC 5.3 was written for a world where a paralegal might draft a contract, file a motion, or interview a client without attorney supervision. The underlying concern was always the same: the attorney is professionally responsible for the work product, even when a non-lawyer produced it.

Add AI to that chain and the risk profile changes in a specific way. AI tools can produce output that looks authoritative, reads fluently, and contains errors that neither the non-lawyer nor the distracted supervising attorney will catch. A paralegal who has always worked carefully and accurately may over-rely on AI output precisely because it sounds confident. The attorney who trusts that paralegal may skip the verification step she would have applied to a human-drafted document.

That's the gap. RPC 5.3 requires you to close it.

What "Reasonable Efforts" Looks Like in 2025

The NJ Supreme Court Committee on Attorney Ethics (ACPE) has not yet issued AI-specific guidance on RPC 5.3 as of this writing, but the structure of the rule gives us enough to work with.

"Reasonable efforts" in a supervised AI context means at minimum three things, and none of them are optional.

First, you need to know what tools your staff is using. This sounds obvious, but it isn't. Paralegals and legal assistants routinely use consumer-grade AI tools, ChatGPT, Claude, Gemini, without telling the supervising attorney. If a staff member is pasting client facts into an unvetted public AI tool, that's a confidentiality problem under RPC 1.6 and a supervision problem under RPC 5.3 simultaneously. You cannot supervise conduct you don't know is happening.

A written AI use policy for non-lawyer staff isn't bureaucracy. It's the documented evidence that you made reasonable efforts. Without it, you're arguing after the fact that supervision was reasonable, and that's a harder position to defend before a disciplinary panel.

Second, you need to define what AI-assisted work requires re-review, not just review. There's a difference between a partner reading an AI-drafted letter with fresh eyes and a partner skimming it expecting it to be fine. The supervision obligation under RPC 5.3 requires the former, particularly when the work touches a legal conclusion, a client advisement, or a filing. Build that into how you structure approvals. If your non-lawyer flags a document as "AI-assisted," that flag should trigger a different level of scrutiny than a document drafted entirely from scratch.

Third, provide actual training, not just instructions. Telling a paralegal "don't use AI for client documents" without explaining why, or what the risks are, satisfies neither the letter nor the spirit of RPC 5.3. Reasonable supervision of AI-assisted work includes helping staff understand what AI tools get wrong and how to spot it. Citation fabrication, confident but incorrect legal summaries, outdated statutory language, paralegals should know what to look for, and that knowledge has to come from somewhere. In a small firm, it comes from you.

The Malpractice Angle Nobody Mentions

Disciplinary exposure under RPC 5.3 is one concern. The malpractice exposure is the other, and in some ways it's more immediate for a solo or two-attorney firm.

If a non-lawyer uses an AI tool to produce a document, the AI output contains a material error, the supervising attorney doesn't catch it, and a client is harmed, the professional liability carrier is going to ask what the supervision policy was. "We trusted our paralegal" is not a satisfying answer when the paralegal was using a tool the firm never evaluated, trained around, or documented.

NJ malpractice insurers are starting to ask about AI use in renewal questionnaires. That alone is a signal worth taking seriously.

A Practical First Step

If you have non-lawyer staff and you have not yet had an explicit conversation about which AI tools they're permitted to use, for which tasks, and what review protocol applies before anything reaches a client, schedule that conversation this week. Document it. Write down whatever you agree to, even informally.

That documentation, imperfect as it may be, is what "reasonable efforts" looks like on paper, and it's far better than nothing when a grievance or a claim forces the question.