Signed, Sealed, Not Quite Delivered: What the NJ Supreme Court's AI Guidelines Still Leave Unanswered for Small Firms

The New Jersey Supreme Court released its "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" on January 24, 2024. At the time, it was among the earliest AI-specific court-level guidance issued by any state supreme court in the country. For NJ practitioners, that timing mattered: it arrived before the ABA weighed in with Formal Opinion 512 (July 2024), before most state bars had even formed an AI working group, and well before generative AI had become a standard line item in law office budgets.

Eighteen months later, the guidelines still read as preliminary, because they are. The Court used that word deliberately. What has changed is that many solo attorneys and small firms have now been using these tools for a full cycle of matters: cases opened, briefed, argued, and closed with AI assistance. The gap between what the guidelines address and what practitioners actually encounter has become more visible.

Here is what that gap looks like in practice.

Where the guidelines give you a clear answer

The Court's January 2024 document grounds its guidance squarely in the existing RPCs. It reminds attorneys that competence under NJ RPC 1.1 includes understanding the tools they use, that confidentiality obligations under NJ RPC 1.6 apply regardless of whether client information passes through an AI system, and that candor to the tribunal under NJ RPC 3.3 is not suspended because a machine drafted the sentence containing the misstatement.

Those anchors are genuinely useful. If you are ever asked to defend a workflow decision, pointing to the Court's own framing of the applicable RPCs is more credible than citing a law blog or a vendor's white paper. The guidelines function as the Court's clearest signal, so far, that AI use is neither categorically permitted nor forbidden. That is actually a meaningful holding for small firms that were waiting for permission they were never going to get.

The document also explicitly states that attorneys remain responsible for the work product regardless of how it was generated. That single sentence closes the argument some practitioners were quietly hoping to leave open.

Where the guidelines go quiet

The guidelines do not define what "understanding" AI tools actually requires. RPC 1.1's competence standard is well-established in the abstract, but the Court did not attempt to specify whether a practitioner must, for example, know how a large language model constructs output, audit vendor security terms, or maintain any documentation of AI-assisted work. All of those are reasonable readings of what competence could require. The guidelines leave each one open.

They also do not address the fee side at all. If AI cuts the hours needed to produce a motion brief from six to two, the question of how to bill that work is live and consequential. The guidelines are silent. NJ RPC 1.5 governs reasonableness of fees, but the Court made no effort to connect that rule to AI efficiency gains.

There is also no guidance on what, if anything, an attorney must tell a client when AI was used to generate advice or draft documents. NJ RPC 1.4 requires keeping clients reasonably informed, but the guidelines do not attempt to translate that into any affirmative disclosure obligation specific to AI. Whether that silence is intentional or just preliminary is unclear.

What "preliminary" actually signals

The Court chose that word carefully. Preliminary guidelines from a state supreme court are not the same as a formal rule amendment or a final advisory opinion from the Advisory Committee on Professional Ethics. They carry genuine weight because the Court issued them, but they do not carry the procedural standing of a rule change under the New Jersey Court Rules.

The ACPE, which handles formal ethics opinions in New Jersey, has not issued an AI-specific opinion as of this writing. ACPE Opinion 701 (2006), which addresses electronic storage and cloud-based data, remains the closest binding guidance on third-party data handling, but it predates generative AI entirely and addresses a narrower question.

What this means practically is that NJ practitioners are being asked to comply with standards that are still being written, using existing rules drafted before the technology existed. That is not a complaint, it is a description of where the profession is.

The one concrete step worth taking now

If you have not read the January 24, 2024 guidelines yourself, do that before anything else. Not a summary, not a CLE slide deck, the document itself. It is short. The Court's language is precise, and the specific RPCs it cites are the ones you will be measured against. When the ACPE eventually issues a formal AI opinion, it will almost certainly build on this foundation. Knowing the foundation lets you read that future opinion critically rather than just accepting someone else's interpretation of it.

What the rule requires

The NJ Supreme Court's "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" (January 24, 2024) identify NJ RPC 1.1 (competence), NJ RPC 1.6 (confidentiality), and NJ RPC 3.3 (candor) as the primary rules governing attorney use of AI tools. The guidelines state that attorneys bear full responsibility for AI-generated work product. They are described by the Court as preliminary. ACPE Opinion 701 (2006) addresses attorney obligations when storing client data with third-party electronic vendors. No NJ court rule or ACPE opinion currently specifies AI-related disclosure obligations to clients or addresses billing adjustments for AI efficiency gains.

What I recommend

Read the January 2024 guidelines in full, not a paraphrase. Map each RPC they cite to a specific decision point in your current AI workflow: which tools receive client data (RPC 1.6), how you verify AI output before filing (RPC 3.3), and what you actually understand about each tool you use (RPC 1.1). Where the guidelines are silent, treat the underlying RPC as your standard and document the judgment call you made. When the ACPE does publish a formal AI opinion, that documentation will show you were operating in good faith against the best available guidance, not guessing.