5 Gaps in NJ's Preliminary AI Guidelines That Solo Attorneys Should Pressure-Test Before Relying on Them

New Jersey's Supreme Court released its "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" on January 24, 2024. If you practice here as a solo or in a small firm, you've probably read them once, maybe nodded, and moved on. That's understandable. At first glance, the document reads as a sensible, high-level framework: verify AI output, protect client data, don't mislead tribunals.

But "preliminary" is doing a lot of work in that title. The Court explicitly described the guidelines as a starting point, not a finished rulebook, and it shows. There are real operational gaps in the document that practitioners are quietly filling with guesswork, which is exactly where disciplinary exposure tends to grow.

Below are five gaps that are worth thinking through carefully before you rely on the guidelines as a compliance baseline.

---

1. The guidelines don't define what counts as adequate verification

The Preliminary Guidelines require attorneys to review and verify AI-generated content before using it. That duty flows directly from NJ RPC 1.1, which requires competent representation including the legal knowledge, skill, thoroughness, and preparation necessary for the representation. But the guidelines don't describe what sufficient verification actually looks like for different task types.

Checking a hallucinated case citation is one kind of verification. Confirming that an AI-drafted contract clause accurately reflects your client's negotiating position is a completely different exercise. A solo attorney reviewing a 40-page AI-drafted will does not have the same verification workload as one running a quick AI summary of a deposition transcript. The guidelines treat these as equivalent by staying silent on the distinction. You'll need to define your own verification standard for each workflow category you operate.

2. The scope of "AI" is left entirely open

The Preliminary Guidelines don't define what tools are covered. Is a grammar-checking plugin covered? A legal research platform that uses AI ranking to surface results? A document management system with AI-assisted OCR? Spell-check has used machine learning for years.

This isn't a trivial line-drawing problem. A small firm that interprets "AI" narrowly and treats only large language model tools as covered could be applying the guidelines to a fraction of its actual AI footprint. Until the Court publishes a revised or final version with a working definition, you should assume the guidelines apply broadly to any tool that uses statistical inference to generate, classify, rank, or summarize legal content.

3. There's no guidance on third-party vendors your client never agreed to

NJ RPC 1.6(a) prohibits revealing information relating to the representation of a client unless the client consents or an exception applies. The Preliminary Guidelines acknowledge the confidentiality concern in general terms but don't specify what disclosures, if any, an attorney must make to a client before running that client's documents through a third-party AI platform.

ACPE Opinion 701, issued in 2006 and covering electronic storage of client files, established that attorneys must take reasonable steps to ensure a third-party provider maintains appropriate confidentiality protections. The same logic applies here, but the Preliminary Guidelines don't extend or update Opinion 701's framework for generative AI contexts. The practical question, specifically whether informed consent is required before uploading client materials to a tool like a cloud-based AI contract reviewer, is unresolved. Conservative practice says you get consent or you anonymize the data first. The guidelines won't back you up either way.

4. They say nothing about non-client AI use

The Preliminary Guidelines are framed around the representation of clients. But solo attorneys use AI for tasks that aren't directly client-facing: drafting firm policies, generating marketing content, analyzing their own business financials, preparing for CLEs. None of that is regulated by the guidelines as written.

That's fine in theory, but it creates a mental model problem. Attorneys who internalize the guidelines as their complete AI framework may stop critically evaluating AI outputs the moment a task feels "internal." Mistakes in internal documents have a way of becoming external problems. The guidelines don't cause this risk, but their narrow framing can.

5. The "ongoing" nature of competence creates a duty with no defined ceiling

The Preliminary Guidelines describe the competence obligation as ongoing, meaning an attorney can't simply learn AI once and check the box. NJ RPC 1.1 supports this reading. But the guidelines give no benchmark for how frequently an attorney must reassess, what triggers reassessment (a new tool version? a new practice area application? a published ethics opinion?), or what continuing education satisfies the duty.

This matters practically because generative AI tools change faster than most attorneys can track. A workflow you validated in early 2024 may be materially different today if the underlying model has been updated or retrained. The guidelines acknowledge the technology is evolving but offer no mechanism for keeping pace with it.

---

The honest read of the Preliminary Guidelines is that they were written quickly, deliberately, and with the expectation that the Court or the ACPE would fill in the details over time. As of today, the ACPE has not issued an AI-specific opinion, so the gaps above remain open.

What the rule requires

NJ RPC 1.1 requires competent representation, including the legal knowledge, skill, thoroughness, and preparation reasonably necessary for the representation. The NJ Supreme Court's Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence (January 24, 2024) require attorneys to review and verify AI-generated content, protect client confidentiality when using AI, and recognize that the competence obligation is ongoing as the technology evolves. ACPE Opinion 701 (2006) requires attorneys to take reasonable measures to ensure that third-party electronic storage providers maintain appropriate confidentiality protections for client data.

What I recommend

Write your own internal definition of "AI" before you rely on the Preliminary Guidelines as a ceiling for compliance. Err broad. For each AI workflow category in your practice (research, drafting, summarizing, intake), document what verification looks like and how long it takes. Treat that documentation as your defense record if a disciplinary question ever arises. Get client consent, or anonymize client data, before uploading materials to any third-party AI platform, because the gap between RPC 1.6 and Opinion 701 is not a safe harbor. And set a calendar reminder to revisit every AI workflow you're running at least twice a year, not because the guidelines say so, but because the tools you validated last year may not be the tools you're actually running today.