What Happens When NJ's "Preliminary" AI Guidelines Are Replaced? How Small Firms Should Plan for the Next Version
Photo by Colin Lloyd on Unsplash
6 minOctober 20, 2026

What Happens When NJ's "Preliminary" AI Guidelines Are Replaced? How Small Firms Should Plan for the Next Version

NJ Supreme Court AI GuidelinesNew Jersey Legal EthicsAI Law Firm Policy

Drafted by AI, reviewed by Adam Elias. Commentary, not legal advice. Published October 20, 2026.

The word "preliminary" is doing a lot of work in the title of New Jersey's Supreme Court document on lawyer AI use. The January 24, 2024 "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" were always intended as a first pass, a signal to practitioners that the Court was watching, not a final statement of what the rules require. That design is both a feature and a problem for solo attorneys trying to build any kind of lasting process around AI.

Here's the practical issue: firms that treated the Preliminary Guidelines as settled law built their workflows on sand. Firms that ignored them entirely are already behind. The right posture is somewhere in between, and it requires understanding what the Preliminary Guidelines actually say, what they were designed to accomplish, and where they are almost certain to change.

What the Preliminary Guidelines are, structurally

The NJ Supreme Court's Preliminary Guidelines do not amend the Rules of Professional Conduct. They interpret existing RPCs in the context of AI, drawing on duties that were already there: competence under RPC 1.1, confidentiality under RPC 1.6, supervision under RPC 5.3. The Court did not create new liability; it applied old liability to new facts.

That framing matters because it tells you what a revised version will likely do. The Court will have more data by the time it issues anything permanent, more reported discipline, more specific AI use cases from actual practice, and possibly guidance from other state supreme courts that moved faster. The final document will almost certainly be more specific, not less.

The ACPE has not issued any AI-specific opinion as of today. ACPE Opinion 701 (2006), which addresses electronic storage and cloud services, remains the closest analog for confidentiality and data storage questions, but it was written before large language models existed as a commercial product. Any practitioner relying solely on Opinion 701 for AI guidance is working with tools that weren't designed for this context.

Where revision is likely to add specificity

A few areas in the Preliminary Guidelines are deliberately vague because the Court didn't have enough information in early 2024 to be more precise.

Disclosure to tribunals. The Preliminary Guidelines note that lawyers have duties of candor under RPC 3.3, but they don't specify when, or in what form, a lawyer must disclose that AI assisted in preparing a submission. Several courts elsewhere have adopted local rules requiring such disclosure. New Jersey has not yet done so at the statewide level, though individual assignment judges have issued their own standing orders. Watch for the revised guidelines or a new court rule to close that gap.

Vendor evaluation standards. The Preliminary Guidelines are silent on what a competent lawyer must actually verify before sending client data to an AI platform. Opinion 701 requires reasonable inquiry into a vendor's data security practices for cloud storage, but that standard was written for cloud file storage, not for tools that process your client's information to generate legal analysis. Expect a revised document to be more prescriptive here.

Non-lawyer use of AI. The Preliminary Guidelines touch on supervision under RPC 5.3(b), which requires a lawyer with supervisory authority to make reasonable efforts to ensure the supervised person's conduct is compatible with the lawyer's professional obligations. But the guidelines don't address the specific question of a paralegal or legal assistant running an AI tool without the supervising attorney ever seeing the output before it influences the work product. That is a gap the Court is likely to close.

What to do between now and the next version

The most important thing a solo or small-firm practitioner can do is build workflows that are rule-anchored, not guideline-anchored. The RPCs themselves aren't going anywhere. A workflow that satisfies RPC 1.6 on confidentiality, RPC 1.1 on competence, and RPC 5.3 on supervision will survive whatever the Court issues next because it's built on the underlying rule, not the guidance document interpreting it.

Practically, that means two things. First, document your AI decisions now. When you decide to use a particular tool for a particular task, write down why you concluded it was appropriate, what you checked about the vendor's data practices, and what verification step you took before relying on the output. That documentation is your defense if discipline is ever at issue, regardless of what version of the guidelines is in effect.

Second, set a calendar reminder to check the NJ Supreme Court's AI resources page every quarter. The Court has not announced a timeline for revised guidelines, but it has signaled that the preliminary document is a living one. Treating it as fixed is a planning error.

The Preliminary Guidelines are a reasonable starting point. They're just not a destination.

What the rule requires

The NJ Supreme Court's Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence (January 24, 2024) interpret existing RPCs, including RPC 1.1 (competence), RPC 1.6 (confidentiality), and RPC 5.3(b) (supervision of non-lawyers), as applying to lawyer use of AI tools. ACPE Opinion 701 (2006) requires lawyers to conduct reasonable inquiry into the security practices of vendors used for electronic client data storage. The Preliminary Guidelines do not amend the RPCs and are explicitly labeled preliminary, signaling that the Court intends to revise them.

What I recommend

Build every AI workflow against the underlying RPC, not the Preliminary Guidelines text. Document your vendor evaluation decisions and your output-verification steps at the time you make them, not after the fact. Check the NJ Supreme Court's AI resources page at least quarterly for any update, standing order, or new court rule on AI disclosure. Don't assume that because the guidelines haven't changed, the Court's expectations haven't either: the revised version will almost certainly be more specific, and firms with documented, rule-anchored practices will be in a far better position to comply quickly when it arrives.

Sources

Read the source before you rely on anything above. The section headed "What the rule requires" states only what these sources require; "What I recommend" is Adam's advice, not the rule.

Get the weekly roundup

New AI Sidebar articles delivered to your inbox. No spam, unsubscribe anytime.