Supervising a Non-Lawyer Who Uses AI: What NJ RPC 5.3 Actually Requires of Small Firm Attorneys

Your paralegal has started using ChatGPT to draft client intake summaries. Your legal assistant is running contract clauses through an AI tool to flag missing provisions. Nobody told them to stop, so they kept going. And now, somewhere in your file system, there are AI-generated work products you've never reviewed and may have forwarded to clients.

This is the scenario NJ RPC 5.3 was built for, and most small firm attorneys aren't treating it that way.

What RPC 5.3 Actually Says

RPC 5.3 places two distinct obligations on attorneys who supervise non-lawyers. First, a lawyer with direct supervisory authority over a non-lawyer must make reasonable efforts to ensure that person's conduct is compatible with the attorney's own professional obligations. Second, if a non-lawyer's conduct would violate the RPCs if done by a lawyer, the supervising attorney is personally responsible for that conduct, either because they ordered it, or because they knew about it and failed to act.

That second prong is the one that matters here. When a paralegal produces an AI-generated document summary with a hallucinated fact, and you send it to a client without checking it, that's an RPC 5.3 problem layered on top of a potential RPC 1.1 competence problem. The fact that the paralegal used a tool you didn't select and didn't personally understand doesn't insulate you.

The New Jersey Supreme Court's approach to attorney responsibility in the supervision context has consistently been that ignorance of a subordinate's method is not a defense. Delegating a task doesn't mean delegating the ethical obligation attached to it.

The Gap in Most Small Firm AI Policies

Here's what typically happens in small firms when AI adoption starts: the attorney starts using an AI tool, finds it useful, mentions it to staff, and staff start using it independently. There's no formal authorization, no defined scope, and no review protocol built around the fact that AI output is probabilistic, not reliable.

That informal adoption pattern creates a gap that RPC 5.3 is designed to close. The rule doesn't require that non-lawyers be prohibited from using AI. It requires that their use of AI be something you've thought through, defined, and are actively supervising.

In practical terms, that means a few things:

Authorized tools only. Your staff should be using AI tools you've reviewed, not whatever they found on their own. This matters both for confidentiality (under RPC 1.6) and for your supervisory accountability. If you don't know what tool they're using, you can't assess whether the output is trustworthy or whether client data is being sent to a third-party server.

Output review is not optional. Any AI-assisted work product that will go to a client, opposing counsel, or a court needs attorney eyes before it leaves the office. That review should be substantive, not a quick glance. The American Bar Association's Formal Opinion 512 (2024) on generative AI made this point clearly: the supervising attorney is responsible for verifying the accuracy of AI-generated content used in legal work.

Document what the review looked like. If something goes wrong later, the question will be whether you had a reasonable supervision system in place. A quick note in a file or a consistent internal workflow showing that attorney review happened before client delivery is the kind of evidence that makes that answer yes.

A Practical Supervision Framework for NJ Small Firms

You don't need a 20-page policy document to be in compliance with RPC 5.3. You need a short, written internal protocol that covers four things:

1. Which AI tools non-lawyers are permitted to use, and for which tasks
2. What categories of information they may not input into any AI system (client names, case numbers, confidential facts)
3. What the review workflow looks like before AI-assisted work product reaches anyone outside the firm
4. Who to ask before using a new AI tool or expanding an existing tool's use

That last point often gets skipped. Staff don't always know that adding a browser extension or trying a new "AI assistant" that auto-processes emails is a material change. Building in a simple ask-first culture closes that loop before it becomes a supervision failure.

Why This Is Particularly Acute in NJ Solo Practices

Solo attorneys face the sharpest version of this problem. In a small firm with multiple partners, supervisory responsibility can be shared and documented across a team. A solo practitioner is the only licensed professional in the room, which means every non-lawyer they employ is exclusively supervised by them. There's no fallback.

If you are a solo attorney in New Jersey with one or two staff members who have started using AI tools informally, the time to formalize that arrangement is before a problem surfaces in a client file. The New Jersey Office of Attorney Ethics has shown, through published discipline, that systemic supervision failures, especially ones involving client harm, draw serious scrutiny regardless of firm size.

A short written protocol, a defined list of approved tools, and a consistent pre-delivery review habit won't guarantee perfection. But they're what "reasonable efforts" looks like under RPC 5.3, and right now, most small firms in New Jersey don't have any of the three.