Supervision Is Not Enough: What the NJ Supreme Court's AI Guidelines Quietly Demand From Managing Attorneys

Most NJ attorneys who heard about the Supreme Court's January 2024 AI guidance filed it away under "be careful with AI" and moved on. That's understandable. The document is titled "Preliminary Guidelines," which sounds tentative, and it arrived during a period when every bar body in the country was issuing something similar. Easy to skim, easy to deprioritize.

That's a mistake worth correcting now, before a grievance or a sanctions motion forces the correction for you.

The NJ Supreme Court's "Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence" (January 24, 2024) ties AI obligations directly to existing RPCs. It doesn't create new rules so much as it maps the existing ones onto AI use, which means the consequences for non-compliance are the same as any other RPC violation. This isn't advisory in the way a bar association newsletter is advisory. The Court issued it, and the Court's disciplinary apparatus enforces the RPCs it cites.

What the guidelines actually say about firm-level responsibility

The Preliminary Guidelines are explicit that the supervising attorney bears responsibility for AI-assisted work product, not just the attorney who ran the prompt. This matters enormously for managing partners and principals at small NJ firms who assume their associates or paralegals are handling AI tasks competently without any formal verification process.

The guidelines cite NJ RPC 5.1(a), which requires partners and managing attorneys to make reasonable efforts to ensure the firm has measures in place giving reasonable assurance that all lawyers conform to the RPCs. They also invoke NJ RPC 5.3(b), which places the same obligation on supervising attorneys with respect to non-lawyer staff. If a paralegal is running AI drafts, the supervising attorney is on the hook for those outputs in the same way they'd be on the hook for any other work product that paralegal produced.

Saying "I reviewed it" satisfies that obligation only if the review was meaningful. A quick scroll is not a review when the underlying task was generating a legal argument, summarizing case law, or drafting a motion. The guidelines make clear that competent review of AI output requires understanding enough about how the tool works to catch the categories of error it's prone to, including confident-sounding citations that don't exist.

The competence requirement runs in both directions

NJ RPC 1.1 requires competent representation, which the Preliminary Guidelines interpret to include understanding the limitations of AI tools in use at the firm. That means a managing attorney who deploys an AI tool across the firm but hasn't evaluated what that tool does with confidential data, how it generates text, or where it tends to fail is already in a difficult position under the rule, even before anything goes wrong.

The guidelines don't specify which tools are acceptable or what training is sufficient. The ACPE has not issued an AI-specific opinion as of this writing, so there's no formal guidance from that body to supplement the Court's document. What the Preliminary Guidelines do say is that the duty to stay current with technology is part of competence, and that this duty applies to AI just as it has historically applied to legal research platforms, e-filing systems, and electronic discovery tools.

For a solo practitioner, this is a personal obligation with no delegation option. For a firm with two or three attorneys, the managing partner likely carries it for everyone.

The disclosure question the guidelines leave open

The Preliminary Guidelines note that existing duties of candor and communication apply when AI is used, specifically pointing to NJ RPC 3.3 (candor toward the tribunal) and NJ RPC 1.4 (communication with clients). The document stops short of mandating blanket disclosure to clients every time AI is used. But it also doesn't give attorneys a clean pass to stay silent.

The functional standard the guidelines suggest is this: if the AI's role in producing work product is material, meaning the AI didn't just fix a typo but actually generated analysis, argument, or advice, then the attorney needs to think carefully about whether silence on that point is consistent with the client relationship and with candor obligations to the court.

That's a judgment call every time, and the Preliminary Guidelines won't make it for you. What they do is put the judgment squarely on the supervising attorney's shoulders.

A few courts are watching closely

It's worth noting that at least some NJ courts are already asking about AI use in filed documents, and the Preliminary Guidelines give them authority to do so. The guidelines explicitly contemplate court-specific rules emerging over time. A small firm that has no internal process for flagging AI-assisted filings is one judicial inquiry away from an uncomfortable situation.

---

What the rule requires

NJ RPC 5.1(a) requires partners and managers to make reasonable efforts to ensure the firm has measures giving reasonable assurance that all lawyers comply with the RPCs. NJ RPC 5.3(b) imposes the same obligation on supervising attorneys with respect to non-lawyer staff conduct. NJ RPC 1.1 requires competent representation, which the NJ Supreme Court's Preliminary Guidelines on New Jersey Lawyers' Use of Artificial Intelligence (January 24, 2024) interpret to include understanding AI tools' limitations. NJ RPC 3.3 requires candor to the tribunal; NJ RPC 1.4 requires adequate client communication. The Preliminary Guidelines apply all of these to AI-assisted practice without creating carve-outs.

What I recommend

Start by actually reading the Preliminary Guidelines in full, they run about eight pages and are available directly from njcourts.gov. Then do a practical audit: list every AI tool currently in use at your firm (including tools staff are using informally), identify who supervises the output from each, and write down what "review" actually means for each task type. A checklist that says "attorney reviewed" is not a process. A checklist that says "attorney verified all cited cases exist in Westlaw or Lexis before filing" is closer to one. If you're a managing partner, share that audit with every attorney and paralegal who touches AI. The Preliminary Guidelines won't protect you if the person running the prompts never saw them.