Can a NJ Attorney Use AI to Screen for Unauthorized Practice of Law Under RPC 5.5?

There's a category of risk that NJ solo attorneys tend to underestimate when they add AI to their intake and client communication workflows. It's not confidentiality. It's not billing. It's whether the AI, or the non-lawyer staff using it, is quietly practicing law on the firm's behalf without anyone noticing.

RPC 5.5 prohibits a lawyer from assisting a person who is not a member of the bar in the unauthorized practice of law. That rule has always applied to paralegals who draft settlement offers without supervision, or legal document services that give specific legal advice. But AI complicates the picture in ways that deserve a concrete look.

Where the AI-Assisted UPL Risk Actually Shows Up

The three most common pressure points in a small NJ firm are intake chatbots, document assembly tools, and AI-drafted follow-up emails sent under a paralegal's name.

Intake chatbots are the most obvious. A solo attorney sets up a chatbot on their website using a general-purpose AI tool, and the bot starts answering prospective client questions. When a user types "Do I have a case?" and the bot responds with anything resembling a legal assessment, that may constitute legal advice to a person who has no attorney-client relationship with the firm. If the chatbot does it autonomously, and the attorney never reviewed the output, the attorney has potentially assisted in UPL under RPC 5.5(a).

Document assembly tools present a subtler version of the same problem. A paralegal pulls a client intake form into an AI tool, asks it to generate a draft letter explaining the client's legal options, and sends it out without attorney review. The letter isn't research. It's not a template. It's legal analysis applied to a specific client's facts, and it went out under conditions a supervising attorney didn't control. RPC 5.5 doesn't require that the non-lawyer intended to practice law. The act is enough.

AI-drafted follow-up emails are where things get genuinely hard to track at volume. When a paralegal uses AI to draft a dozen status update emails and a few of them drift from factual updates into explaining legal strategy or predicting outcomes, each of those is a potential issue. The attorney who put the AI tool in place and delegated email drafting without a review protocol has a problem that RPC 5.5 and RPC 5.3 share ownership of.

What NJ Ethics Guidance Says (and Doesn't Say)

The NJSBA has not yet issued formal guidance specifically addressing AI and UPL, as of mid-2025. That gap is not a green light. NJ courts and the Supreme Court's Disciplinary Review Board have consistently interpreted RPC 5.5 broadly. The rule applies to the attorney who enables the UPL, not just the person committing it.

The ABA's 2023 Formal Opinion 512 on generative AI touched on supervision but didn't resolve the UPL question for AI acting autonomously. What it did confirm is that attorney oversight is the load-bearing wall. If the AI does something that would constitute legal advice when done by a human, the attorney is responsible for the output.

A Practical Screening Protocol for Small NJ Firms

If you're using AI at the intake or client communication layer, the following controls reduce your RPC 5.5 exposure in concrete ways.

First, audit every AI touchpoint where a person who is not yet a client or is a current client receives output. Map those touchpoints on paper, even a simple list. For each one, ask whether the output could be characterized as legal advice specific to that person's facts.

Second, build a bright-line rule into your intake chatbot's system prompt: no legal assessments, no predictions, no specific recommendations. The bot schedules consultations and collects information. That's it. Review the prompt every 60 days because AI tools update their underlying models and behavior can shift.

Third, any AI-drafted client communication that goes out under the firm's name should require a one-line attorney sign-off in your practice management system before delivery. "Reviewed" with initials and a date takes 20 seconds and creates a defensible record. A paralegal sending AI-generated emails without that checkpoint is a RPC 5.5 incident waiting to happen.

Fourth, if you use a document assembly platform that allows non-lawyers to generate client-specific documents, configure it so completed documents are routed to an attorney review queue rather than delivered directly. Most platforms in the NJ market support workflow approvals. Use them.

The Question Worth Asking Yourself

Most NJ solo attorneys who have a UPL exposure in their AI workflow don't know about it because they set the tool up once and moved on. The better practice is to pick one workflow per quarter, trace every step from client data input to output delivery, and ask whether a non-lawyer or automated system is making a call that properly belongs to a licensed attorney. That's not a heavy audit. It's a 30-minute walk-through. And it's the kind of thing that keeps a RPC 5.5 complaint from becoming a referral to the Office of Attorney Ethics.