Training Staff on AI Tools in NJ: What the Supreme Court's Preliminary Guidelines Actually Expect From Solo Attorneys

Solo attorneys often ask me some version of the same question: "I've got a paralegal who's started using ChatGPT to draft client letters. Do I need to do anything formal about that?" The short answer is yes, and the NJ Supreme Court's Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence (January 24, 2024) give you a cleaner framework than most practitioners have actually read.

The guidelines don't create a separate training mandate sitting somewhere in a checklist. Instead, they operate through the RPCs they explicitly invoke, and one of the most practically loaded is RPC 5.3(b), which requires a lawyer with supervisory authority over a non-lawyer to make reasonable efforts to ensure the person's conduct is compatible with the professional obligations of the lawyer.

That one sentence does a lot of work when a paralegal is generating first drafts of demand letters, summarizing deposition transcripts, or pulling statutes using an AI tool you may or may not have vetted.

What "compatible conduct" actually means for AI use

When the Preliminary Guidelines address non-lawyer assistance, they draw a clear line: the attorney remains responsible for the work product, regardless of which tool produced it. The guidelines state that lawyers must take responsibility for AI-generated work and apply the same professional judgment they would to any other output from a supervised person.

That responsibility doesn't dissolve because the staff member, not the attorney, clicked the button. Under RPC 5.3(b), if a paralegal submits an AI-generated brief summary to a client without the attorney reviewing it for accuracy, and it contains a hallucinated case citation or a wrong statute number, that's the attorney's problem. The guidelines reinforce this by tying accuracy review to core competence and diligence obligations.

So what does a "reasonable effort" actually look like? In the context of AI use by staff, I'd argue it has at least three practical components: the attorney must know which AI tools staff are using, must set explicit parameters for how those tools are used in client-facing work, and must build in a review step before output leaves the office.

The training gap most solos are ignoring

Here's where many solo and small firms are exposed: they've informally permitted staff to use AI tools without ever discussing the limits. No one said "you can use this for X but not Y." No one explained that AI output requires verification before it goes into a filing or a client email. And no one documented any of that.

The Preliminary Guidelines don't require a formal written training program, but they do require the attorney to exercise actual supervisory judgment. A hypothetical: suppose a paralegal in a two-person family law firm uses a general-purpose AI to draft a letter advising a client on the tax implications of a proposed marital settlement. The paralegal doesn't know the AI's training data has a cutoff that predates a relevant IRS rule change. The attorney never asked what the paralegal was using or reviewed the output. The client acts on the letter. Under RPC 5.3(b) as the Preliminary Guidelines read it, that attorney has a real exposure problem.

Informal permission is not the same as reasonable supervisory effort.

What the guidelines say about verification

One of the Preliminary Guidelines' clearest directives is that lawyers must verify AI-generated content before using it in any context involving clients or tribunals. The guidelines specifically call out the risk of inaccurate output, including fabricated legal authority, and they place the verification duty squarely on the attorney.

When a staff member is the one generating that content, the verification duty doesn't transfer to the staff member. It stays with the supervising attorney. That means building a workflow where staff-generated AI output is flagged as such and reviewed before it's finalized, not just spot-checked occasionally.

This doesn't require expensive software. A shared intake log where staff note "AI-assisted draft" is enough to cue a review step. What matters is that the step exists and happens consistently.

What the rule requires

Under NJ RPC 5.3(b), a supervising attorney must make reasonable efforts to ensure that non-lawyer staff conduct is compatible with the lawyer's professional obligations. The NJ Supreme Court's Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence (January 24, 2024) require attorneys to verify AI-generated content before use, to take responsibility for AI-assisted work product, and to apply professional judgment to any output produced with AI assistance, whether by the attorney or by supervised staff.

What I recommend

Start with a direct conversation with every staff member who touches client files, even informally, about which AI tools they're using and for what. Document that conversation, even in a simple email. Then set explicit limits in writing: what tasks are permitted with AI assistance, what the review step looks like, and who signs off before output reaches a client or a court. If you don't know what tools your staff are using, that's the first thing to fix, because you can't supervise what you can't see.