Unauthorized Practice of Law or Competitive Edge? What NJ RPC 5.5 Actually Says About AI Doing Legal Work in Your Firm

There is a version of this story that ends with a grievance.

A solo immigration attorney in a mid-sized New Jersey firm deploys an AI chatbot on her website. It asks intake questions, assesses visa eligibility, and — because the prompt engineering was a little too ambitious — tells a prospective client which visa category they should apply for. The attorney never sees that exchange before the client acts on it. Six months later, the application is denied, and the client is asking why they weren't told about a disqualifying factor the AI never flagged.

Was that unauthorized practice of law? Was it something worse?

This is not a hypothetical designed to scare you. It is a pattern of risk that is quietly taking shape inside New Jersey small firms right now, and NJ RPC 5.5 — the rule governing unauthorized practice of law — is the ethical boundary that defines it.

---

What RPC 5.5 Actually Governs (and Why AI Complicates It)

RPC 5.5 prohibits a New Jersey attorney from assisting a non-lawyer in the unauthorized practice of law. Traditionally, that meant watching your paralegal or legal assistant. The rule exists because legal judgment — analyzing facts against law, advising a specific person on a specific situation — must be performed by or under the meaningful supervision of a licensed attorney.

AI is not a licensed attorney. It is, functionally, a very capable non-lawyer. And when an AI tool performs legal analysis, generates advice, or communicates conclusions to a client without adequate attorney oversight, the attorney who deployed it may be facilitating exactly what RPC 5.5 was designed to prevent.

The New Jersey Supreme Court has not issued an AI-specific ethics opinion on this yet. But the NJSBA's ethics guidance and the broader ABA framework make the analytical path clear: the character of the output matters, not the character of the tool producing it. If the output constitutes legal advice, someone licensed must be responsible for it.

---

Three Workflows Where NJ Firms Are Quietly Crossing the Line

1. Autonomous client intake chatbots. These tools are genuinely useful for gathering basic information — name, matter type, dates. The risk appears when the chatbot is configured to interpret facts and return a preliminary legal assessment. "Based on what you've described, you may have a viable personal injury claim" is legal advice. If no attorney reviewed that response before delivery, you have a supervision problem that RPC 5.5 cannot easily absorb.

2. AI-generated client-facing documents sent without attorney review. Some firms use document automation tools that generate retainer agreements, demand letters, or cease-and-desist notices triggered by client-entered data — and then email them directly to clients or counterparties. If the attorney's role is limited to setting up the template once, months ago, that is not meaningful supervision of a specific legal matter. It is delegation without oversight.

3. Staff-operated AI research tools producing client-ready memos. A paralegal runs a legal research query through an AI platform, edits the formatting, and emails the output to the client as a "research summary." The attorney is copied but never substantively reviews the analysis before it goes out. This is a supervision failure under both RPC 5.3 and RPC 5.5 — the non-lawyer is effectively delivering legal work product without attorney gatekeeping.

---

What Meaningful Supervision of AI Actually Looks Like

The good news is that none of the above scenarios require you to abandon AI. They require you to design workflows with deliberate attorney checkpoints. Here is what that looks like in practice:

• Chatbots should collect, not conclude. Configure intake tools to gather facts and schedule consultations — never to render assessments, even qualified ones. Add a disclosure that no attorney-client relationship is formed and no legal advice is given through the intake interface.

• Documents require a review gate before transmission. Any AI-generated document going to a client or third party should pass through an attorney's active review — not just a checkbox, but a substantive read. Build this into your project management workflow as a mandatory step, not an optional one.

• Staff AI use needs a defined approval chain. If your paralegals or legal assistants use AI tools to generate research or draft correspondence, establish a written protocol that requires attorney sign-off before any output leaves the firm. Document that review in the matter file.

• Audit your prompts periodically. The instructions you give your AI tools — your system prompts, your templates — function like standing orders to a member of your staff. Review them quarterly to ensure they are not inadvertently instructing the tool to perform legal analysis autonomously.

---

The Practical Bottom Line for NJ Solo Attorneys

The question RPC 5.5 asks about your AI use is simple, even if the answer requires some thought: is a licensed New Jersey attorney exercising meaningful, contemporaneous judgment over the legal work before it reaches the client?

Meaningful does not mean ceremonial. Reviewing an AI-generated letter for typos while the underlying legal analysis goes unchecked is not supervision — it is rubber-stamping. New Jersey's disciplinary system has never looked kindly on that distinction in the non-lawyer context, and there is no reason to expect it will treat AI differently.

The attorneys who will use AI most effectively — and most safely — are those who treat it as a capable junior contributor that still requires an experienced senior review before anything goes out the door. That is not an obstacle to efficiency. That is what competent law practice has always looked like.

The tool changes. The responsibility does not.