Picking Apart the NJ Supreme Court's AI Guidelines: What a Solo Practitioner Actually Owes a Client Before Running Any AI Tool

Most solo attorneys in New Jersey have skimmed the Supreme Court's "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" (issued January 24, 2024) and filed them mentally under "be careful with AI." That reading is understandable. It's also incomplete in ways that can create real professional exposure.

The Guidelines aren't a checklist. They're closer to a framework that maps existing RPC duties onto a new category of tool. And when you read them closely, they demand something specific before you open a tab and start prompting: you have to understand what the tool does well enough to supervise it and protect your client.

That's a pre-use standard, and most solo practitioners haven't built a habit around it.

What "understanding the tool" actually means under the Guidelines

The NJ Supreme Court's Preliminary Guidelines identify competence, confidentiality, supervision, and candor as the four pillars AI use touches. On competence, they're explicit: a lawyer must understand the benefits and limitations of a given AI tool before using it on client work. The competence obligation comes from NJ RPC 1.1, which requires the legal knowledge, skill, thoroughness, and preparation reasonably necessary for the representation.

Reading those two documents together, "using AI competently" isn't just about checking output for errors after the fact. It includes understanding, going in, where the tool is likely to fail. A general-purpose large language model hallucinates. A legal-specific tool trained on a narrow corpus may produce confident answers that are months out of date. A summarization tool may drop qualifying language that changes an analysis entirely. If you don't know which failure mode applies to the tool you're running, the Guidelines suggest you're not yet in a position to use it on a client matter.

For a solo attorney without an IT department or a supervising partner, this lands squarely on you.

The supervision gap the Guidelines quietly expose

NJ RPC 5.3(b) 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. The NJ Supreme Court's Preliminary Guidelines apply this duty explicitly to AI tools, treating the output of an AI system as something requiring the same supervisory attention a lawyer would give a paralegal's work product.

The practical implication for solos is uncomfortable: if you're using AI to draft a motion, summarize a deposition, or pull together case law, you aren't just a user. You're the supervisor. The Guidelines don't create an exception for small firms. They assume the solo attorney fills every role, including the one responsible for checking the work.

That means a quick read-through isn't supervision. Supervision, in the RPC 5.3 sense the Guidelines invoke, requires knowing what errors to look for, having a method to catch them, and being able to correct them before they reach a client or a court.

Before the first prompt: a realistic pre-use standard

Hypothetically, consider a solo family law attorney who uses a general AI chatbot to draft initial settlement proposals. The tool is fluent, fast, and confident. The attorney reviews the output but doesn't know the tool's training cutoff, doesn't know whether it was fine-tuned on legal content or general text, and doesn't know whether it's retaining client inputs for model training. Under the Guidelines, that attorney has a problem even if the output is correct this time, because the duty of competence requires understanding the limitations before relying on the output.

A workable pre-use habit looks something like this: before you use any AI tool on a client matter, spend twenty minutes with the tool's published documentation and terms of service. You're looking for three things specifically: (1) what the tool was trained on and when, so you know its knowledge boundaries; (2) how it handles data you input, because NJ RPC 1.6 confidentiality doesn't pause for convenience; and (3) what the tool's own documentation says about accuracy and limitations, because vendors frequently bury disclaimers that matter professionally.

This isn't a one-time exercise. When a vendor updates its model or its terms, you repeat it. The Guidelines are preliminary, but NJ RPC 1.1 is not.

Where the Guidelines are silent, and why that matters

The Preliminary Guidelines don't specify which tools are acceptable, don't set a disclosure threshold for telling clients when AI was used, and don't address pricing or billing for AI-assisted work. The Supreme Court said explicitly that the January 2024 document is preliminary, and that more guidance will follow as the technology and case law develop.

That silence isn't permission to fill the gaps however you like. The existing RPCs still apply. But it does mean that solos in New Jersey are currently operating without bright-line rules on several practical questions, and that the disciplinary exposure in those gaps is real if your use of a tool later comes under scrutiny. The safer posture, professionally and practically, is to err toward more documentation of your pre-use review, not less.

What the rule requires

NJ RPC 1.1 requires the legal knowledge, skill, thoroughness, and preparation reasonably necessary for a representation. NJ RPC 5.3(b) requires a supervising lawyer to make reasonable efforts to ensure non-lawyer 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) apply both duties directly to AI use, requiring lawyers to understand the benefits and limitations of any AI tool before using it on a client matter, and to supervise AI output with the same diligence applied to non-lawyer work product.

What I recommend

Before using any AI tool on a client matter, document your pre-use review in a short internal note: what the tool is, what you confirmed about its training data and cutoff date, how it handles input data, and what its documented limitations are. Keep that note in the file. If the matter ever faces scrutiny, that note is evidence that your use of AI met the competence and supervision standard the Guidelines describe. Update the note when you switch tools or when a vendor updates its model. Twenty minutes of documentation now is considerably less painful than reconstructing your reasoning later.