What NJ's Preliminary AI Guidelines Get Wrong About Small Firms (And What They Get Right)

When the New Jersey Supreme Court issued its "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" on January 24, 2024, practitioners across the state reacted with the usual mix of relief and anxiety. Relief because something official had finally come out. Anxiety because "preliminary" is not exactly a confidence-inspiring modifier when your bar license is on the line.

Having read the document carefully, my honest assessment is this: the guidelines work well as a reminder of existing duties, but they leave solo and small-firm practitioners with real unanswered questions that the existing RPCs don't cleanly resolve.

What the guidelines actually say

The Supreme Court's preliminary guidelines don't create new rules. They map existing obligations onto AI-assisted practice. They remind practitioners that NJ RPC 1.1 (competence) requires understanding the tools you use, that NJ RPC 1.6 (confidentiality) applies when client data is fed into AI systems, and that lawyers remain responsible for any work product generated with AI assistance. The guidelines explicitly state that lawyers must review AI output and cannot simply adopt it as their own without verification.

That's sound, and it tracks the same analytical framework as ABA Formal Opinion 512 (July 29, 2024), which the ABA issued several months later. Both documents treat AI as a tool that sits inside the existing professional responsibility structure, not outside it.

The guidelines also touch on fees, flagging that billing for redundant AI-assisted work that an efficient attorney would have done faster may implicate RPC 1.5. And they acknowledge the supervision question: if a non-lawyer at your firm is running the AI, your duties under RPC 5.3 don't disappear.

Where the document goes thin for small firms

Here's where I think the guidelines underserve solo and small-firm practitioners specifically.

The document was written at a level of generality that made sense for a Supreme Court pronouncement directed at the entire bar. But most of the harder operational questions show up at the small-firm level, not at AmLaw 100 firms with dedicated IT and risk management teams. A solo family law attorney in Middlesex County using a consumer AI tool to draft a first pass at a property settlement agreement faces a very different risk calculus than a fifty-attorney firm using an enterprise legal AI platform with a BAA in place.

The guidelines don't distinguish between those situations. They don't tell you how to evaluate whether a specific tool's data retention policy is compatible with RPC 1.6. They don't address what "sufficient review" looks like when you're the only attorney and also the only reviewer. They don't explain whether disclosing AI use to a client is mandatory under RPC 1.4 or discretionary. That last question, in particular, is one that practitioners keep asking, and the preliminary guidelines leave it genuinely open.

The guidelines also acknowledge their own incompleteness by calling themselves "preliminary." That word is doing a lot of work. It signals that more formal guidance is likely coming, but it gives practitioners no timeline and no indication of what form that guidance will take.

The ACPE's silence matters here

It's worth being precise about one institutional point: as of this writing, the NJ Advisory Committee on Professional Ethics (ACPE) has not issued an AI-specific opinion. The Supreme Court's preliminary guidelines are the only New Jersey-specific authority we have. ACPE Opinion 701 (2006), which addressed cloud storage and electronic data generally, is still the closest the Committee has come to addressing data-transmission questions that arise in AI use, but it predates generative AI by nearly two decades and maps imperfectly onto the current tools.

That gap matters. Ethics opinions carry practical interpretive weight that a set of guidelines, preliminary or otherwise, does not fully replicate. An ACPE opinion on generative AI use, confidentiality, and disclosure would give small-firm practitioners something concrete to build workflows around.

The most practical thing you can do right now

Given what we have, the most defensible position for a NJ solo or small-firm attorney is to treat the Supreme Court's preliminary guidelines as a floor, not a ceiling. Document your AI review process for every matter where you use AI-generated content. Keep a record of which tool you used, what input you provided, and what verification steps you took before filing or sending anything. If your tool's terms of service permit the vendor to use your inputs for model training, that's a RPC 1.6 question you need to answer before the next file goes in.

The guidelines won't protect you if something goes wrong and you can't show you thought through the risk. The documentation will.

What the rule requires

The NJ Supreme Court's "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" (January 24, 2024) state that lawyers must understand the AI tools they use (consistent with RPC 1.1), must protect client confidentiality when using AI (RPC 1.6), must review and take responsibility for all AI-generated work product, and must ensure that fees for AI-assisted work are reasonable under RPC 1.5. The guidelines do not create new standalone rules; they apply existing RPCs to AI-assisted practice. The ACPE has issued no AI-specific opinion. ACPE Opinion 701 (2006) addresses electronic storage and cloud services generally under RPC 1.6.

What I recommend

For NJ solo and small-firm attorneys, I recommend four concrete steps. First, read the January 2024 preliminary guidelines directly, not through summaries. Second, create a short written log for any matter where AI generates substantive content: tool name, date, input type, and your verification method. Third, review your AI vendor's data retention and training terms against the confidentiality standard in RPC 1.6 before using the tool on client matters. Fourth, watch for a formal ACPE opinion on AI, which I expect to be the document that finally answers the disclosure and confidentiality questions the preliminary guidelines leave open. When that opinion comes out, your existing log will make it easier to assess whether your current workflow already complies.