How NJ Attorneys Should Actually Read the Supreme Court's Preliminary AI Guidelines Before Using Any Tool

Most attorneys who've read the New Jersey Supreme Court's "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" (January 24, 2024) came away with a general impression: verify AI output, protect client data, be careful. That reading isn't wrong, but it's incomplete in ways that matter.

The Guidelines aren't self-contained. They operate by reference to the RPCs, which means understanding what they actually demand requires going back to the underlying rules. Skipping that step leaves solo attorneys with a vague sense of duty but no concrete framework for what happens when something goes wrong.

Here's how to read the document with more precision.

The Guidelines don't create new obligations; they map existing ones

The most common misreading is treating the Guidelines as a standalone regime. They're not. The document explicitly situates AI use within existing professional conduct rules, including NJ RPC 1.1 (competence), NJ RPC 1.4 (communication), NJ RPC 1.6 (confidentiality), and NJ RPC 5.3 (responsibilities regarding non-lawyer assistance). The Guidelines' primary function is to explain how those rules apply when the "assistant" is a software tool rather than a paralegal.

That framing has a practical consequence: the standard you're held to is the same standard you've always faced. The question isn't "did I follow the AI guidelines?" It's "did I satisfy my obligations under the RPCs, given that I used an AI tool?" A solo attorney who runs a contract through an AI drafting tool and files the result without review hasn't just "not followed the guidelines." They've potentially breached RPC 1.1's duty of competence and RPC 3.3's duty of candor to the tribunal, depending on what the AI got wrong.

The word "preliminary" is doing real work

The document's title labels these as preliminary guidelines. That word signals something specific about how the Court intends them to be used. Preliminary guidance typically means the issuing body is watching how the technology develops and reserving the right to impose stricter or more specific requirements later.

For a solo practitioner, this creates a planning obligation that isn't spelled out anywhere in the text. If you build a workflow around the current guidelines and the Court issues a revised version with, say, specific disclosure requirements or file-retention rules, your current setup may need to change on short notice. The prudent move is to document your current AI practices now, not just comply with them, so any future delta is easy to identify and close.

What the guidelines say about verification, specifically

One section of the Guidelines addresses accuracy and verification. The Court states that attorneys must review AI-generated work product for accuracy and completeness, applying the same professional judgment they'd bring to any work product. This isn't a suggestion to "double-check things." It's a restatement of the supervisory standard under RPC 5.3(b), which requires a supervising attorney to take reasonable remedial action when they know, or should know, that a non-lawyer's conduct would constitute a violation if done by an attorney.

Practically, that means a one-pass review isn't automatically sufficient. If your AI tool hallucinates a citation or misconstrues a statutory term, and a reasonable attorney reviewing the output should have caught it, the Guidelines (read through RPC 5.3) say the error is yours. The sophistication of the tool doesn't change the analysis.

Confidentiality and the gap the Guidelines leave open

On data handling, the Guidelines point attorneys toward their obligations under RPC 1.6 but don't specify what technical or contractual safeguards satisfy those obligations. That gap is significant. ACPE Opinion 701 (2006), issued long before generative AI existed, addressed electronic storage of client files and established that attorneys must use reasonable care to prevent unauthorized access, including reviewing a vendor's terms of service and data security practices.

The Guidelines don't supersede Opinion 701. They coexist with it. A New Jersey attorney using a cloud-based AI drafting tool is still subject to the Opinion 701 framework for the underlying client data, even if the AI use itself is governed by the 2024 Guidelines. That means the same due diligence obligations on vendor terms and data handling apply.

What the rule requires

The NJ Supreme Court's Preliminary Guidelines on NJ Lawyers' Use of Artificial Intelligence (January 24, 2024) require attorneys to apply professional judgment when using AI, verify AI-generated output for accuracy and completeness, and satisfy their obligations under the existing RPCs, including RPC 1.1 (competence), RPC 1.6 (confidentiality), and RPC 5.3(b) (supervision of non-lawyer assistance). ACPE Opinion 701 (2006) separately requires attorneys to use reasonable care to prevent unauthorized access to client data stored or processed electronically, including reviewing vendor data security practices.

What I recommend

Read the Guidelines once more, but this time with the RPC text open in a second tab. For each paragraph, identify which underlying rule it's invoking and what that rule actually requires. You'll find the Guidelines are thinner than they look and that the real substance is in the RPCs themselves.

Document your current AI workflow in a one-page internal memo: what tools you use, what tasks they perform, how you review their output, and what vendor terms you've accepted. Treat it as a baseline you can update when the Court issues revised guidelines, because it will. Store that memo with your engagement letter templates so it's easy to find, not in a folder you'll forget about in six months.