State Bar AI Guidance: Every Verified Ethics Opinion, State by State

There is no national rule governing lawyers' use of generative AI. ABA Formal Opinion 512 (July 29, 2024) is the closest thing to a baseline, and at least 16 U.S. jurisdictions — 15 states plus the District of Columbia — have issued their own formal ethics opinions or adopted bar guidance as of August 2026. Every row in the table below was verified at the issuing bar, court, or regulator. That matters more than it should: AI-compiled trackers keep citing a Utah opinion that appears nowhere in that bar's published opinions.

Is there a national rule for lawyers using AI?

No, and there will not be one soon. The ABA's Formal Opinion 512 interprets the Model Rules, which bind no one; each state's adopted rules of professional conduct do. What Opinion 512 supplies is the analytical frame nearly every state now uses: generative AI implicates competence (Rule 1.1), confidentiality (1.6), communication (1.4), candor (3.3), supervision (5.1 and 5.3), and reasonable fees (1.5). No opinion anywhere prohibits the technology. Every opinion everywhere holds the lawyer responsible for what it produces.

From that baseline, the states split into three tiers: those with formal ethics opinions or published guidance (most of the table below); those that adopted task-force reports rather than numbered opinions (New York and Minnesota, both in the table because their reports were formally adopted); and those that have said nothing at all, which does not mean nothing applies. Trackers that also count judicial standing orders, pending committees, and CLE materials reach twenty-five jurisdictions or more; this table counts only what we could confirm at a primary source, so treat it as a verified floor, not a ceiling. A fourth tier is forming: at the California Supreme Court's direction, the State Bar has drafted amendments to the Rules of Professional Conduct themselves, out for public comment in 2026 — which would make AI duties enforceable rule text for the first time anywhere.

Which states have AI ethics opinions for lawyers?

Sixteen jurisdictions, verified, in the order they acted. Each link goes to the opinion itself or the issuing body's own page, not to a secondhand summary.

Two reading notes. First, the issuing body varies, and it matters: New Jersey's guidance came from its Supreme Court, Missouri's from the court's ethics counsel, West Virginia's from its Lawyer Disciplinary Board, and Virginia's opinion carries its Supreme Court's approval, so in those jurisdictions the guidance sits closer to enforcement than an advisory committee's view. Second, this list is a snapshot; more opinions are in committee, and the count will grow. What will not change is the method: believe the opinion, not the tracker, and click through to the primary source before you rely on a row.

What do state bar AI opinions require?

Read them all and the same five duties surface every time. The differences are emphasis, not direction.

Competence and verification. Universal: the lawyer must understand what the tool does and must verify what it produces. D.C.'s Opinion 388 devotes more space to hallucination risk than any other; Pennsylvania's 2024-200 requires checking every citation; Texas Opinion 705 says lawyers cannot blindly rely on generated answers; Alabama calls blind acceptance of AI work a competence failure outright. Kentucky's E-457 runs the duty in the other direction too — a lawyer who refuses available AI may one day fail the competence standard. Verification is also the one duty software can meaningfully shoulder: CaseRead's free Hallucination Shield checks each citation in a draft (up to 25 per run) against real court data across all 53 U.S. jurisdictions. No tool discharges the duty itself; the lawyer signing the filing does.

Confidentiality. Every opinion applies Rule 1.6 to the prompt box. California and North Carolina warn against client information in consumer tools; Missouri requires reading the terms of service before anything goes in; Florida recommends informed consent before a third-party tool sees confidential information; Alaska tells lawyers to anonymize their prompts when a tool cannot promise protection. The operative question in every version: where does the input go, who can read it, and does it train the model? We break that analysis down in Can I upload client documents to ChatGPT? — and it is why CaseRead isolates each firm's documents in the firm's own database schema and never uses client data for training.

Client disclosure and consent. The genuine split. West Virginia demands written informed consent, full stop. Pennsylvania requires communication about AI use and consent in certain circumstances. New Jersey requires disclosure when the client asks or when informed decision-making depends on it. Alabama expects candor with clients about how AI figures in their work. Kentucky finds no duty at all for routine use. If you practice in more than one of these states, your engagement letter should satisfy the strictest one you touch.

Billing honesty. Near-universal: you may not bill hourly for time AI saved. California, Florida, D.C., Texas, and Alaska all say it plainly; Kentucky nudges lawyers to pass the efficiency on; Virginia adds the useful counterpoint that a flat or value-based fee need not drop just because the work went faster. The full analysis, including what to do about AI line-item costs, is in Billing for AI-assisted work.

Supervision. West Virginia and Missouri make the analogy explicit: treat the model like a nonlawyer assistant under Rules 5.1 and 5.3. Florida, Pennsylvania, North Carolina, Texas, and Alaska land in the same place — firm policies, training, and a lawyer answerable for what the tool did. It can draft; it cannot be responsible.

What should lawyers do in states with no AI guidance?

Practice as if your state had issued the consensus opinion — because functionally it has. The silent states did not suspend their rules of professional conduct, and every opinion in the table reaches its holdings by applying rules your state already adopted in nearly identical form. Competence, confidentiality, candor, supervision, and honest fees do not need an AI-specific opinion to bind you, and Opinion 512 shows exactly how a disciplinary authority is likely to apply them.

Concretely: verify every AI-assisted citation before filing — courts sanction fabricated citations under existing rules in every state, as the case list in whether lawyers can use AI for legal research shows; keep client-identifying information out of consumer tools; bill actual time; and write down how your firm does all three. A one-page written policy is the highest-leverage move for a firm in a silent state, because it converts the consensus of the table above into something your staff actually follows. Our law firm AI policy guide includes a template built on exactly these themes.

Does Utah have an AI ethics opinion?

No — and Utah is the sharpest illustration of why this table only includes what we verified. Several AI-assembled trackers cite a "Utah State Bar Ethics Opinion 24-03" on generative AI. No such opinion appears in the Bar's published ethics opinions. A phantom citation, circulating in trackers about verification duties, is exactly the failure this post exists to correct.

What Utah actually has is arguably more instructive than an opinion. The Utah Artificial Intelligence Policy Act (S.B. 149, effective May 1, 2024, and narrowed by 2025 amendments) — the first generative-AI consumer-protection statute in the country — requires a person in a state-licensed occupation, lawyers included, to disclose when a consumer is interacting with generative AI directly in a high-risk service. And the Utah Court of Appeals supplied the enforcement lesson in Garner v. Kadince, 2025 UT App 80, sanctioning attorneys whose petition cited ChatGPT-fabricated cases: opposing counsel's fees, a full client refund, and a $1,000 donation to legal aid. A statute plus a sanctions order adds up to a stricter posture than several states with formal opinions. Our full deep-dive, Utah AI ethics for lawyers, walks through the rules, the statute, and the case, and runs down the phantom opinion in detail.

The map will change; the duties won't

More opinions are coming — Alabama's arrived in 2026 — and California may soon move the duties into rule text. But the practical instruction has been stable since Michigan and California went first in the fall of 2023: understand the tool, protect the client's information, verify the output, bill the truth, and stay responsible for the work. That is what CaseRead is built around: the AI associate that knows your matter, searches your firm's files and the public law together, and cites only what it verified. Run your next brief through the free Hallucination Shield, and if you want the verification built into the research itself, start with the free tier.

Frequently asked questions

Which states have AI ethics opinions for lawyers? At least 16 U.S. jurisdictions have issued formal generative-AI ethics opinions or adopted bar guidance as of August 2026: Alabama, Alaska, California, Florida, Kentucky, Michigan, Minnesota, Missouri, New Jersey, New York, North Carolina, Pennsylvania (jointly with the Philadelphia Bar), Texas, Virginia, West Virginia, and the District of Columbia. Counts vary because some trackers also include task forces and court standing orders. Every entry in this list traces to the issuing bar, court, or regulator.

Is ABA Formal Opinion 512 binding on state bars? No. ABA Formal Opinion 512, issued July 29, 2024, interprets the ABA Model Rules, which have no force of their own — only a state's adopted rules of professional conduct bind its lawyers. In practice, Opinion 512 is the national baseline: most state opinions track its analysis of competence, confidentiality, communication, candor, supervision, and fees, and disciplinary authorities in silent states are likely to find it persuasive.

What do state bars require for AI confidentiality? Every opinion applies the duty of confidentiality to AI inputs: before client information goes into a tool, the lawyer must know where it goes, who can see it, and whether it trains the model. California's guidance says not to input confidential information into tools that lack adequate protections, Missouri requires vetting the platform's terms of service, North Carolina disfavors client data in publicly available tools, and West Virginia goes furthest — informed client consent confirmed in writing before generative AI touches the matter.

Do lawyers have to tell clients they use AI? It depends on the state; this is where the opinions genuinely split. West Virginia requires informed consent confirmed in writing. Pennsylvania's joint opinion calls for communicating about AI use and obtaining consent in certain circumstances. New Jersey requires disclosure if the client asks or cannot make an informed decision without knowing. Kentucky finds no duty to disclose routine use, and Virginia's fee opinion does not reach the question. Check your jurisdiction's opinion before assuming either answer.

Does Utah have an AI ethics opinion? No. As of August 2026, no formal generative-AI ethics opinion appears in the Utah State Bar's published opinions, though AI-generated trackers keep citing a 'Utah Ethics Opinion 24-03.' Utah lawyers are governed by the existing Rules of Professional Conduct, the Utah AI Policy Act's disclosure duty for consumer-facing AI in high-risk regulated services, and the warning of Garner v. Kadince, 2025 UT App 80, where the Court of Appeals sanctioned attorneys over ChatGPT-fabricated citations.