When a NJ Associate's AI Research Memo Went Wrong — What RPC 5.3 Requires of the Supervising Attorney

Picture this: your paralegal uses a legal AI tool to draft a case law summary for a motion brief. You're swamped, you trust her work, and the memo looks clean. You sign off, the brief gets filed. Two weeks later, opposing counsel flags a cited case that doesn't say what the memo claimed it said.

Under New Jersey's Rules of Professional Conduct, that's not your paralegal's problem. It's yours.

RPC 5.3 Is the Rule Small Firms Most Casually Ignore

New Jersey RPC 5.3 — Responsibilities Regarding Non-Lawyer Assistants — requires lawyers with direct supervisory authority over non-lawyer staff to make reasonable efforts to ensure those individuals' conduct is compatible with the professional obligations of the attorney. For partners or managing attorneys in small firms, this obligation extends to the entire operation.

In the age of AI, this rule has a sharper edge than it did five years ago. When a paralegal ran a Westlaw search, their output was bounded by what Westlaw actually indexed. When a paralegal runs a prompt through a general-purpose or even purpose-built legal AI tool, the output can be fluent, authoritative-sounding, and completely fabricated — with no visible warning sign.

The supervising attorney cannot outsource responsibility to the tool. And critically, they cannot outsource it to the non-lawyer using the tool.

What "Reasonable Efforts" Actually Means in an AI Workflow

The phrase "reasonable efforts" in RPC 5.3 has always been contextual — courts and ethics boards look at the systems and processes a firm had in place, not just whether a specific error occurred. That framing becomes operationally important when AI is involved, because it means a disciplinary inquiry won't just ask what went wrong. It will ask what safeguards you had in place to catch it.

For NJ solo and small-firm practitioners, reasonable efforts under an AI-integrated workflow should include, at minimum:

• A written prompt-to-review protocol. Non-lawyers using AI tools should follow a defined process: what prompts are permissible, what jurisdictions and databases the tool draws from, and how output is flagged before it reaches an attorney.
• Mandatory verification checkpoints. Any AI-generated legal research must be verified against primary sources before it is incorporated into a client deliverable or court filing. This isn't optional — it's the substantive equivalent of cite-checking.
• Documented supervisory touchpoints. If a staff member used AI to produce work product, that fact should be noted in the file with a record of attorney review. In a disciplinary proceeding, "I reviewed it" is far weaker than a timestamped review notation in your practice management system.
• Tool-specific training, not general AI literacy. There's a meaningful difference between teaching a paralegal what AI is and training them on the specific tool your firm uses — its known limitations, its citation behavior, and the document types where hallucination risk is highest.

The Supervision Gap in NJ Small Firms

Here's the pattern I see repeatedly in small and solo practices: the attorney adopts an AI tool, hands access to a paralegal or legal assistant to save time, and never formalizes how that tool fits into the supervisory structure. There's no policy, no training record, no review protocol. The tool becomes part of the workflow quietly, organically, and invisibly.

That informality is fine until it isn't.

The New Jersey Office of Attorney Ethics has not yet issued AI-specific formal guidance under RPC 5.3 — but the rule text doesn't require a new opinion to apply. The obligation to supervise non-lawyers was written to be technology-neutral. When the New Jersey Supreme Court's District Ethics Committees have examined supervision failures in past cases, they've consistently looked at whether the attorney created and maintained adequate oversight structures. The introduction of AI tools doesn't change that analysis; it raises the stakes.

A Practical Restructuring for Firms Using AI With Non-Lawyer Staff

If a paralegal or legal assistant in your firm currently uses any AI tool — even something as routine as a grammar checker with legal features, or a template drafting assistant — the following structure isn't optional, it's defensive:

1. Name the supervising attorney for AI-assisted work product in your office policy. In a solo practice, that's you. In a two- or three-attorney firm, it should be explicit.
2. Require AI-generated research to carry a "draft/unverified" label until an attorney has independently checked the citations.
3. Add AI tool usage to your non-lawyer onboarding checklist. If a staff member hasn't been trained on the specific tool's limitations, they shouldn't be using it on client matters.
4. Build a one-line notation habit in your file system. "Research memo drafted with [Tool]. Citations verified by [Attorney] on [Date]." That line may matter more than you think if a grievance is ever filed.

RPC 5.3 isn't a new concern. But AI has fundamentally changed the surface area of what non-lawyers can produce — and how confidently wrong that output can be. The supervisory architecture your firm had in 2019 was not designed for this. It's worth thirty minutes to rebuild it for 2025.