State bar AI guidance: a 50-state survey

The current state of generative AI regulation in the U.S. legal profession is a patchwork. Roughly half of the state bars have issued formal ethics opinions or guidance documents on lawyers' use of generative AI. The other half have not. The federal framework — ABA Formal Opinion 512, issued July 29, 2024 — sets the baseline, but bar associations are not federal regulators, and the actual professional discipline a lawyer faces operates at the state level.

For a lawyer practicing in one state, the rules are usually findable in one or two documents. For a lawyer practicing across jurisdictions — and increasingly, for any firm whose work touches federal court, multi-state class actions, or out-of-state clients — the picture is genuinely complicated.

This article maps the landscape. Which states have issued formal guidance. What the major opinions say. Where the guidance diverges. And what every multi-jurisdiction practice needs to do in light of all of it.

The federal baseline

Before walking through the states, the federal baseline matters because every state bar's analysis sits on top of it.

ABA Formal Opinion 512 identified six duties that every lawyer using generative AI owes under the Model Rules of Professional Conduct: competence (Rule 1.1), confidentiality (Rule 1.6), communication (Rule 1.4), candor toward the tribunal (Rule 3.3), supervision (Rules 5.1 and 5.3), and reasonable fees (Rule 1.5).

These are Model Rules. They are not directly binding on any lawyer. The Model Rules have been adopted, in substantially similar form, by every U.S. jurisdiction except California — and even California's Rules of Professional Conduct, while structurally different, address the same underlying duties. The practical effect: ABA Formal Opinion 512's framework is the framework most state bars are working from, even when they have not formally adopted its conclusions.

State guidance, where it exists, typically does one of three things: (1) confirms that the Model Rules duties apply to AI use; (2) adds state-specific procedural requirements (like court-filing disclosure requirements); or (3) clarifies how a particular duty applies in the AI context based on the state's own rules of professional conduct.

The duties don't go away just because a state hasn't issued specific guidance. They still apply through the underlying rules.

The states with comprehensive formal opinions

Twelve states have issued formal ethics opinions specifically addressing generative AI as of mid-2026. These are the documents that carry the most weight — they are written by the same committees that produce binding ethics interpretations on every other professional responsibility issue, and they typically include both the analytical framework and concrete examples.

California — first out, most comprehensive. The State Bar of California issued Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law on November 16, 2023, predating ABA Opinion 512 by eight months. The California guidance was updated by the Board of Trustees on May 14, 2026 to address the use of agentic AI and provide additional clarifications. It is the most detailed state-level guidance currently in circulation. Notable features: explicit treatment of confidentiality in third-party AI services, detailed handling of supervision duties for AI outputs, and direct language on the duty to verify citations against authoritative sources.

Florida — early formal opinion. The Florida Bar Ethics Opinion 24-1 was issued on January 19, 2024 — six months before the ABA's federal opinion. The Florida opinion identifies four core ethical concerns: confidentiality, competence/accuracy, billing practices, and lawyer advertising restrictions. The advertising angle is unusual; Florida treats client-facing AI chatbots as advertising and imposes requirements that the chatbot must identify itself as AI (not a human) and that any claims about AI capabilities must be objectively verifiable. Florida also treats generative AI tools as analogous to non-lawyer assistants for Rule 5.3 supervision purposes.

New Jersey — Supreme Court guidance. The New Jersey Supreme Court issued AI guidance to its lawyers via the Committee on Attorney Advertising, the Advisory Committee on Professional Ethics, and the Committee on the Unauthorized Practice of Law in January 2024. The New Jersey approach emphasizes that existing ethics rules apply to AI use without requiring new rules — lawyers must verify accuracy, maintain confidentiality, supervise outputs, and bill reasonably.

District of Columbia — DC Bar Ethics Opinion 388. The DC Bar's Legal Ethics Committee issued Ethics Opinion 388 in April 2024, covering attorneys' use of generative AI in client matters. The DC opinion is notable for its detailed treatment of confidentiality in the context of cloud-based AI services and its specific guidance on engagement letter language.

North Carolina — 2024 Formal Ethics Opinion 1. The North Carolina State Bar issued 2024 Formal Ethics Opinion 1 addressing both extractive and generative AI. The North Carolina opinion is distinctive for its attempt to draw lines between AI uses that are part of standard practice (research databases that use AI to surface results) and those that require special handling (generative AI that produces draftable content). The supervision discussion is particularly detailed.

Pennsylvania — Joint Formal Opinion 2024-200. Pennsylvania's joint opinion from the Pennsylvania Bar Association's Committee on Legal Ethics and Professional Responsibility and the Philadelphia Bar Association's Professional Guidance Committee covers the major ethical issues plus a Pennsylvania-specific item: requirements around disclosure of AI use in court submissions. Pennsylvania practitioners need to attend to local court rules about whether and when AI use must be disclosed in a filing.

Kentucky — KBA E-457. The Kentucky Bar Association issued ethics opinion KBA E-457 covering generative AI. The Kentucky guidance follows the ABA framework closely with state-specific procedural notes.

Michigan — AI FAQ format. Michigan's State Bar published Artificial Intelligence for Attorneys — Frequently Asked Questions rather than a numbered formal opinion. The FAQ format is approachable but less formally binding; it walks through specific scenarios attorneys are likely to encounter.

Texas — Opinion No. 705. The Texas State Bar's Professional Ethics Committee issued Opinion No. 705 in February 2025 providing specific guidance on lawyers' use of generative AI. Texas's opinion addresses both the duties under the Texas Disciplinary Rules of Professional Conduct and Texas-specific concerns about client trust account handling when AI services are paid through firm operating accounts.

Utah — "Using ChatGPT in Our Practices." The Utah State Bar published guidance specifically framed around ChatGPT and similar tools. The Utah guidance is notable for its direct treatment of the Mata v. Avianca failure mode and its emphasis on verification.

New Mexico — Bar Opinion 2024-004. The State Bar of New Mexico issued Opinion 2024-004 on September 24, 2024. New Mexico's opinion is particularly detailed on the supervision angle and on the duty to verify AI-generated work product.

Virginia — informal guidance via Virginia State Bar. Virginia has issued informal guidance through the Standing Committee on Legal Ethics. The Virginia approach emphasizes that existing rules adequately address AI use without specific new requirements.

The states with task force reports or non-binding studies

A second group of states has produced reports, task force studies, or informal recommendations that don't have the formal status of an ethics opinion but still carry weight as the state bar's expressed position.

New York — NYSBA Task Force Report (April 6, 2024). The New York State Bar Association's Task Force on Artificial Intelligence produced a comprehensive 85-page report adopted by the NYSBA House of Delegates on April 6, 2024. The report is unusual in scope — it covers ethical issues, but also broader questions of AI's impact on the profession, on access to justice, and on professional licensure. New York has not, as of mid-2026, issued a formal ethics opinion from the state's professional responsibility apparatus, but the Task Force Report functions as the de facto guidance.

Georgia — Special Committee on AI and Technology. Georgia has formed a special committee studying AI; guidance is in development.

Illinois — Committee recommendations. The Illinois State Bar has provided initial committee recommendations but has not adopted a formal ethics opinion.

Iowa — Resource list. The Iowa State Bar Association has compiled a list of AI resources for member education rather than issued formal guidance.

Tennessee — Task force exploring recommendations. A task force is studying the issues; formal guidance is in development.

Washington — Legal technology task force. Washington's task force is studying AI as part of broader legal technology issues; formal guidance is in development.

Hawaii — Committee exploring recommendations. A committee is studying AI use; guidance is in development.

The states without formal guidance

The remaining states have not, as of mid-2026, issued formal AI-specific guidance. This list includes (but is not limited to): Alabama, Alaska, Arizona, Arkansas, Colorado, Connecticut, Delaware, Idaho, Indiana, Kansas, Louisiana, Maine, Maryland, Massachusetts, Minnesota, Mississippi, Missouri, Montana, Nebraska, Nevada, New Hampshire, North Dakota, Ohio, Oklahoma, Oregon, Rhode Island, South Carolina, South Dakota, Vermont, West Virginia, Wisconsin, and Wyoming.

The absence of state-specific guidance does not mean the duties don't apply. Every one of these states has adopted Rules of Professional Conduct based on the ABA Model Rules. The competence, confidentiality, candor, supervision, and reasonable fees duties apply through those rules. ABA Formal Opinion 512, while not binding, is highly persuasive interpretive authority in any state whose rules track the Model Rules — which is essentially every U.S. state.

A lawyer in Wyoming who uses ChatGPT without verification is not in a different position from a lawyer in California who does the same thing. The lawyer in Wyoming may not have an ethics opinion in front of her specifically addressing AI. She has, instead, the same underlying rules and the same body of federal sanctions cases — Mata v. Avianca, Park v. Kim, and the rest — establishing exactly what the duties require.

Common themes across the guidance

The states that have issued guidance vary in format and emphasis, but the substantive themes converge tightly. Five themes appear in essentially every state opinion that addresses AI.

One — confidentiality is the dominant concern. Every state opinion addresses the fact that consumer AI tools transmit user inputs to provider servers, that those servers may retain or process the inputs, and that this implicates the lawyer's confidentiality duty. The states that go furthest (California, Florida, DC, North Carolina) require lawyers to investigate the AI tool's data retention and use practices before transmitting client information.

Two — competence requires understanding the technology. Every opinion makes explicit that the competence duty includes understanding the AI tool's capabilities and limitations. A lawyer who treats a generative AI tool as a search engine without understanding that it can fabricate authoritative-sounding output is not competent to use it.

Three — the duty to verify is universal. Every state opinion that addresses citation use directly says that generative AI outputs must be independently verified before being relied upon in court filings. The verification duty does not depend on the lawyer's good-faith belief that the output is accurate.

Four — supervision applies to AI outputs the same way it applies to non-lawyer assistant work. Most opinions invoke Rule 5.3 (or the state analog) and treat AI tools as the functional equivalent of a non-lawyer assistant whose work the lawyer must supervise.

Five — billing must reflect actual time and value, not "what it would have taken before AI." The fees discussion is consistent across opinions: lawyers cannot bill at rates that don't reflect the actual time spent, even when AI compresses the work substantially.

Where the guidance diverges

Five themes converge. Three themes diverge meaningfully.

Disclosure to clients. States differ on whether and when lawyers must affirmatively disclose AI use to clients. California's guidance leans toward case-by-case judgment driven by materiality. Pennsylvania has gone further and requires disclosure of AI use in court submissions in certain circumstances. New Jersey emphasizes that existing communication duties already require what's needed. The Florida opinion treats client-facing AI chatbots as advertising and imposes specific disclosure requirements in that context.

Court-filing disclosure. Federal and state courts increasingly impose their own AI disclosure rules at the local-rule level, layered on top of state bar guidance. The Pennsylvania state bar guidance addresses this explicitly. Other states leave it to individual courts. The result is that a lawyer practicing in multiple federal districts may have different AI disclosure obligations in different courthouses — and the state bar guidance may not capture all of them.

Client consent for confidentiality. States differ on what triggers the duty to obtain informed consent before submitting client information to a generative AI tool. California and Florida treat the question carefully and provide nuanced guidance on when consent is needed. Other states take harder lines requiring consent for any consumer AI tool that doesn't have enterprise-grade confidentiality contracts. The North Carolina opinion is one of the more detailed treatments of when consent is and isn't required.

What multi-jurisdiction practitioners need to do

For a firm that practices in more than one state — which describes essentially every firm with federal court practice, every firm with multi-state corporate or class action work, and many regional firms — the patchwork creates a real compliance challenge.

The pragmatic response is to operate at the highest common denominator. Identify the strictest applicable rule in any jurisdiction where the firm practices, and build the firm's AI policies around compliance with that rule. This is the same approach firms have used for years on questions like data retention, conflict-checking, and disciplinary reporting — adopt the strictest rule and you've adopted all of them.

For most firms, that means:

1. Use only AI tools with contractual zero-retention and no-training terms. This satisfies confidentiality requirements in every state that has issued guidance.
2. Adopt a firm-wide written AI policy. Required (or strongly implied) by supervision duties in every jurisdiction.
3. Train every user. Required by competence duties.
4. Build verification into the workflow rather than relying on individual lawyer discretion. Required by the candor duty in every jurisdiction, and the only structurally reliable way to comply with it.
5. Maintain audit logs of AI use. Not explicitly required in every jurisdiction, but required to defend the work against any subsequent challenge — and the only way to demonstrate compliance with the verification duty if it's ever questioned.
6. Address AI use in engagement letters. Required or implied in most jurisdictions' communication guidance; protects against later disputes about client expectations.
7. Bill consistent with actual time expended. Universally required.

Firms that build their AI infrastructure around these seven items are compliant in every U.S. jurisdiction, including the ones that haven't yet issued formal guidance. When those states do issue guidance, it will almost certainly track the existing pattern.

Why we built DDEAS around the strictest reading

When we designed Discover Docket's ethics and accountability framework, we did not try to solve the AI compliance problem at the level of any one state's rules. We solved it at the level of the strictest reading of any state's rules, plus a layer of infrastructure that responds to the next thing the bar associations are going to ask for, before they ask for it.

DDEAS — the Discover Docket Ethical and Accountability Standards — assumes the lawyer is practicing in the jurisdiction with the most demanding AI requirements. It enforces structural verification of every citation before the output reaches the lawyer's screen. It produces audit logs that satisfy every state's documentation requirements. It runs inside infrastructure with contractual confidentiality protections that satisfy the strictest state bar's confidentiality analysis.

A firm using Discover Docket in Wyoming is not operating to Wyoming's current (non-existent) AI guidance. It is operating to California's, which is the most detailed in the country. A firm using Discover Docket in Maine is not waiting for Maine's bar to issue an opinion. It is already compliant with the opinion Maine is likely to issue when it gets to it.

Read the DDEAS framework →

The patchwork is going to keep evolving. New opinions will issue. Existing opinions will be updated. Local court rules will multiply. The lawyer using infrastructure designed for the strictest reading is the lawyer who doesn't have to track all of it.