Generative AI is already being used in professional practice to draft documents, summarize records, generate designs, analyze images, and answer questions. Across many professions, licensing guidance remains limited or has not yet explained how existing professional duties may apply to those uses.

New Jersey, through Assembly Bill A4731, has proposed an unusually broad response. The bill would require the New Jersey Division of Consumer Affairs to establish a model generative AI policy for professions and occupations regulated through its licensing boards. The Director would have six months after enactment to establish a model policy through expedited rulemaking. Each board would then have nine months, or until its next regularly scheduled meeting following establishment of the model, to adopt it with any profession- or occupation-specific adaptations the board deems necessary. Introduced on March 10, 2026, A4731 advanced through the Assembly Science, Innovation and Technology Committee and the Assembly Regulated Professions Committee and is now pending before the Assembly Appropriations Committee. It has not passed either legislative chamber.

The appeal is obvious. One model could give dozens of licensing bodies a common starting point without waiting for dozens of separate legislative projects. The difficulty, however, is equally apparent. The same policy architecture would need to reach physicians, engineers, accountants, pharmacists, psychologists, and many others. A policy general enough to travel that far may say too little. A policy detailed enough to govern one profession may arrive badly angled for the next.

A model policy can establish a common control architecture. It cannot supply the profession-specific judgment that gives the controls meaning. The space between those two things is where A4731 will either become governance or become paperwork. 

Can a Single Policy Fit Every Profession?

The model policy mechanism is the bill’s most interesting feature and also its central design challenge. The structure resembles a model law approach: one common starting point followed by adaptation by more specialized regulatory bodies. Here, however, the adaptations would occur across professions within a single state rather than across states regulating the same industry.

Building AI governance frameworks is harder than it looks. Even with substantial institutional resources and subject matter expertise, getting the policy architecture right for one large organization requires sustained, iterative effort and an honest accounting for contextual risk. The bill does not specify who will draft the model policy, which professionals will participate, or what legal, technical, and operational expertise the Division must consult. For a policy intended to reach professions ranging from personal services to medicine, that omission is not administrative trivia.

The bill’s scope may present its own threshold question. A4731 defines generative AI to include systems that generate nonscripted outputs with “limited or no human oversight.” In professional practice, however, human participation can occur at several different stages. A professional may select the tool, construct the prompt, review and revise the output, and remain responsible for the result even if the system generated its initial response with little human involvement. The Division and boards may therefore need to determine whether “limited” oversight refers to how the system generates an output, how the professional uses that output, or both, before defining the policy's practical scope.

That common architecture will not eliminate the differences among the professions it covers. A physician reviewing AI-assisted image analysis, an engineer approving an AI-generated design, and a pharmacist evaluating AI-assisted medication guidance are not exercising the same judgment, working from the same evidence, or creating the same risks.

A usable model would standardize the architecture, not the professional judgment. Common elements could include tool approval, data restrictions, documentation, disclosure, verification, supervision, incident reporting, and periodic review. The boards would still have to determine when independent verification is mandatory, which uses are prohibited, what records must be retained, how much human review is meaningful, and what professional consequences attach to a failure. 

Where Common Architecture Meets Professional Judgment

Lawyers sit outside A4731’s Division of Consumer Affairs framework, but recent developments in the legal profession offer a useful comparison. California enacted SB 574 on September 30, 2026. Effective January 1, 2027, the law addresses attorneys’ duties concerning nondelegation, confidentiality, verification and correction of AI-assisted materials, disclosure to courts when generative AI is used in documents submitted to them, and personal review of cited authorities.

The State Bar of California has pursued a parallel professional regulation track. It conducted two rounds of public comment in 2026 on proposed AI-related amendments to six Rules of Professional Conduct. The first period closed May 4, and the second closed August 6, 2026. The amendments remain proposed. California’s approach is profession-specific and increasingly detailed. A4731 begins at the opposite end, with a common baseline intended for many professions, followed by board level adaptation.

A4731 does not itself define an AI violation, prescribe a penalty, or establish a common enforcement standard. Those questions would emerge through the model policy, the rules each board adopts, and the statutory and disciplinary authority each board already possesses. That may permit appropriate professional tailoring, but it also creates the possibility that substantially similar AI-related conduct will be framed, investigated, and sanctioned differently across professions.

The rulemaking will therefore need to address several recurring design questions:

• Whether materially similar conduct should produce comparable consequences across boards;

• How boards will assess causation when AI influenced but did not determine an outcome;

• When invocation of “professional judgment” reflects meaningful review and when it conceals its absence; and

• What technical capacity boards will need to investigate alleged violations.

Florida illustrates the alternative; it took a narrower route. Effective June 15, 2026, amended Rule 2.515(d)(2) requires every signer of a court filing, whether counsel or self-represented, to represent that cited legal authorities exist and are accurately cited. The rule responds to documented problems involving fabricated or inaccurately cited authorities, including material generated through AI tools, but regulates the filing and its citations rather than the technology used to prepare it. It supplies a specific representation, an identifiable actor, and express sanctions authority after notice and an opportunity to be heard. A4731 aims for far broader coverage. The tradeoff is that many of the most important operational questions remain for the Division and individual boards.

What Profession-Specific Scaffolding Looks Like

The New Jersey Supreme Court issued preliminary AI use guidelines for lawyers in January 2024. On March 30, 2026, it issued a separate notice encouraging attorneys and firms to develop, adopt, and periodically update AI policies and attached a starter template. The notice emphasized that attorneys remain fully responsible for their work product and professional obligations and identified review, training, confidentiality, and periodic updating as elements of responsible AI use. The attached starter template expressly cautions that adopting a policy is not a “safe harbor.” As the template puts it, “A document alone does not mitigate, remove, or prevent mistakes, confidentiality lapses, or professional consequences.” Careful implementation, training, supervision, and verification remain necessary.

Separately, the New Jersey State Bar Association’s Task Force on Artificial Intelligence and the Law issued a May 2024 report supplying practical materials, including policy development considerations, a sample AI-use policy, and vendor diligence questions.

The State Bar of Georgia has taken a similar practice-focused approach through its Generative AI Toolkit, a resource that connects generative AI use to the Georgia Rules of Professional Conduct and addresses competence, confidentiality, fees, conflicts, candor, supervision, vendor diligence, and human review. The toolkit offers guidance rather than a formal advisory opinion, but it illustrates how a profession can translate existing duties into operational expectations.

The legal profession therefore demonstrates what profession-specific scaffolding can look like when courts and professional institutions have already developed guidance, policy templates, and established ethical duties against which AI use can be evaluated. The harder question is what happens across professions where comparable groundwork does not yet exist. The most consequential issue will be when a professional may rely on AI, when independent verification is required, and what professional judgment cannot be delegated. Those thresholds will vary sharply across disciplines.

What to Watch

Even if it advances no further, A4731 presents an approach other states may consider. If enacted, the first consequential moment will be what the Division produces during the bill’s accelerated initial rulemaking. The second will be whether individual boards treat “profession- or occupation-specific adaptations” as substantive work or as a search and replace exercise.

Practitioners and trade groups should not assume that the initial process will resemble ordinary notice and comment rulemaking. The bill directs the Director and boards to adopt the initial rules notwithstanding the ordinary Administrative Procedure Act requirements. Those initial rules could remain in effect for no more than one year, after which they would have to be adopted, amended, or readopted under the ordinary APA process. Early engagement may therefore matter before a conventional public comment record exists.

A successful model would not pretend that every covered profession presents the same risk profile. It would identify the controls that should travel across professions and require each board to explain where its profession needs something different. The template starts the governance work. The boards will have to finish it.

For questions about AI governance program design, sectoral compliance strategy, standards alignment, or navigating the current AI regulatory landscape, contact the Jones Walker Privacy, Data Strategy, and Artificial Intelligence team. Stay tuned for continued insights from the AI Law and Policy Navigator.

Disclosure: Co-author Jason M. Loring serves as Vice Chair of the State Bar of Georgia’s Artificial Intelligence and Emerging Technology Committee.