AI Agents for Lawyers

Regulation & Ethics

US State Bar AI Ethics Opinions: A 2026 State-by-State Guide

As of June 2026, 17 US jurisdictions plus the ABA have formal AI ethics opinions. None bans AI. They agree on verification, split on client consent and billing.

By Dr. Niklas SchmidtUpdated 13 November 202611 min read
On this page
  1. Who has spoken, and what each opinion is known for
  2. Where every opinion lands in the same place
  3. The three fault lines
  4. California 2026: from guidance to rule text
  5. The outliers and the layers around them
  6. One policy that satisfies all seventeen
  7. What to do this month

As of June 2026, 17 US jurisdictions plus the ABA have issued formal ethics opinions on lawyers' use of generative AI, according to the legalaicompliance.help tracker. None bans the technology. All demand competence, confidentiality and personal verification, and all forbid billing for time the tool saved. They split on two things: whether clients must consent, and what California did in 2026, which was to rewrite the rule comments themselves.

Who has spoken, and what each opinion is known for

The count is less useful than the content. Most opinions restate the six duties in ABA Formal Opinion 512 with local rule numbers. A handful say something the others do not, and those are the ones worth reading in full.

JurisdictionOpinionDateBest known for
FloridaEthics Opinion 24-119 Jan 2024Informed consent before a third-party tool sees confidential information; a client-facing chatbot must say it is not a lawyer; efficiency "must not result in falsely inflated claims of time"
District of ColumbiaEthics Opinion 388Apr 2024Technology does not alter the duties; the lawyer owns the output
ABAFormal Opinion 51229 Jul 2024The six-duty framework every later opinion borrows; informed consent where a self-learning tool may expose client data
New York City BarFormal Opinion 2024-52024Reported to exempt embedded tools such as Word, Westlaw and Lexis from any disclosure
Illinois Supreme CourtAI Policy1 Jan 2025"Disclosure of AI use should not be required in a pleading"; the filer is accountable for the final work product
TexasOpinion 705Feb 2025"A reasonable and current understanding of the technology"; no fee for time "saved"; per-use tool charges may be passed through at cost
OregonFormal Opinion 2025-205Feb 2025Competence with AI; candour with the court is what determines discipline
VirginiaLEO 1901Approved 24 Nov 2025A reasonable value-based fee need not be reduced because AI saved time
New York City BarFormal Opinions 2025-6 and 2026-222 Dec 2025; 5 Aug 2026AI meeting recorders: consent of all parties, no surreptitious recording, default is not to record
AlabamaFormal Opinion 2026-012026The only new numbered opinion of 2026
CaliforniaCOPRAC rule comments; Practical Guidance13 Mar 2026; 14 May 2026First body to address agentic AI and to move AI into the commentary of the rules themselves

The rest of the roll, by date: Kentucky E-457 (15 March 2024), Pennsylvania and Philadelphia Joint Opinion 2024-200 (22 May 2024), West Virginia LEO 24-01 (June 2024), North Carolina 2024 FEO 1 (1 November 2024), Mississippi 267 (14 November 2024), New Mexico 2024-004 (2024), Alaska 2025-1 (23 April 2025) and Washington 2025-05 (2025). These follow the ABA pattern closely; if your state is on this list, read your own opinion once and then work from the union checklist below.

Where every opinion lands in the same place

Strip away the local rule numbers and the opinions converge on four propositions.

Competence means understanding the tool, not just the law. Texas puts it most directly: a lawyer who chooses to use generative AI "must have a reasonable and current understanding of the technology", because only then can the lawyer evaluate hallucination risk, data limitations and the exposure of confidential information. Forty states have adopted Comment 8 to Model Rule 1.1, the duty of technology competence, so this is not a Texas peculiarity. The State Bar of Texas's own 2026 survey of 1,553 lawyers shows the gap: AI use rose from 30% in 2024 to 62% in 2026, yet only 9% said they were "extremely confident" evaluating AI-generated documents.

Confidentiality turns on the tool's terms. Every opinion distinguishes self-learning tools that store and reuse inputs from tools that do not. Texas lists four precautions drawn from its earlier cloud-computing opinion: understand how the technology works, review and if necessary renegotiate the terms of service, learn the data-security protections, and train lawyers and staff. The confidentiality checklist on this site does the vendor-by-vendor work.

Verification is personal and non-delegable. Texas: lawyers "cannot blindly rely upon or use answers given by generative AI tools" and are responsible for the work product "regardless of who (or what) does the original research and drafting". Oregon 2025-205 goes a step further by making candour with the court the variable that determines discipline. The sanctions casebook shows what happens when this proposition is ignored.

Hourly billing cannot include time saved. Texas quotes DC Opinion 93-379 from 1993: when a lawyer bills hourly, "the economies associated with the result must inure to the benefit of the client". Florida says the same. The billing page works through the arithmetic.

The three fault lines

Client consent. Florida 24-1 and ABA 512 call for informed consent before a self-learning tool is exposed to confidential information. Texas 705 asks for "thoughtful analysis", says lawyers "should consider informing clients" and "may need to secure client consent", and stops short of a rule. The New York City Bar's 2024-5 is reported to carve out embedded tools altogether. In practice the line runs between tools: an enterprise tier with contractual no-training terms rarely triggers consent; a consumer account with default training does. The disclosure page sets out engagement-letter wording for both.

Fee structure. No opinion lets you bill hours you did not work. The disagreement is what happens outside hourly billing. Virginia LEO 1901, approved by the Supreme Court of Virginia on 24 November 2025, says a reasonable value-based fee need not shrink because AI made the work faster. California's 2026 Practical Guidance adds two constraints from the cost side: general AI subscriptions are overhead, and AI costs cannot be marked up without informed written consent. Texas allows per-use charges to be passed through, but not above the amount actually paid to the provider.

Disclosure to the court. The Illinois Supreme Court's policy, effective 1 January 2025, states that disclosure "should not be required in a pleading". Chief Justice Mary Jane Theis said the court's "current rules are sufficient to govern its use". Meanwhile more than 30 federal district courts require certification, and two Ohio judges ban generative AI in filings. Bar opinions and court rules are separate systems; the standing orders page covers the second.

California 2026: from guidance to rule text

California is the only jurisdiction that has moved AI out of advisory guidance and into the commentary of its Rules of Professional Conduct, and the only one to address agentic AI by name. The sequence matters because other states will copy it.

DateStepWhat it did
16 Nov 2023Practical Guidance for the Use of Generative AIFirst state-level guidance; advisory
22 Aug 2025California Supreme Court letterDirected the State Bar to consider moving the guidance into the rules and to address agentic tools that "autonomously perform tasks or workflows without human prompting"
13 Mar 2026COPRAC approves AI-specific comments to six rulesCompetence, communication, confidentiality, candour, supervision, non-lawyer assistants; public comment to 4 May 2026
14 May 2026Board of Trustees approves updated Practical GuidanceCovers generative and agentic AI; described as a "living document"
1 Sep 2026Legislature passes a lawyer-AI billReported by Above the Law; Senator Umberg: "if they can't do that, everything breaks down"

Three lines from the 2026 guidance, as summarised by Wilson Elser, do most of the work. First: "No paper filed in court should contain citations that the attorney responsible for the filing has not personally read and verified." Second: general AI subscriptions are overhead, and no mark-up on AI costs is permitted without informed written consent. Third, and the one with the widest reach, the proposed Rule 1.6 comment would treat exposing client data to an AI tool as "revealing" it wherever there is "a material risk of unauthorized access, retention, or use". That reframes the confidentiality question from "did anyone see it" to "could the vendor retain it", which is a question about contract terms.

The agentic point is not academic. When an agent drafts, files or sends without a prompt for each step, the rule that bites is 5.3 on non-lawyer assistants: the lawyer must supervise the agent as if it were a paralegal with no judgment and perfect confidence. The State Bar's public comment page and the ethics and technology resources hold the primary documents. Enforcement has started: the State Bar has charged attorneys over AI-generated fake citations, and one accepted a 30-day suspension plus ten hours of technology CLE, according to the Daily Journal.

The outliers and the layers around them

Alabama is the only jurisdiction to add a numbered opinion in 2026, and it did so in the year its Supreme Court dismissed an entire appeal (24 April 2026) over AI-tainted briefs, ordered $17,200 in fees and barred the lawyer from filing without a co-signing attorney. The court, not the bar, set the pace.

The New York City Bar has drifted from general guidance to specific tools. Opinions 2025-6 and 2026-2 deal with AI recorders and transcription: consent of all parties, no surreptitious recording under Rule 8.4, and a default of not recording. If your firm uses a meeting bot, the meeting notes page applies those rules.

CLE mandates are a separate track. New Jersey requires one technology credit per two-year cycle from 1 January 2027; Florida requires three technology hours per cycle; North Carolina one hour a year; New York one cybersecurity hour a year. A secondary tracker counts over 400 approved AI-related CLE hours across major providers by April 2026, up from about 60 in early 2024, but reports that 78% are general primers and fewer than three courses cover agentic workflows.

Malpractice cover has not yet reacted. No major US lawyers' professional liability carrier had filed an explicit AI exclusion as of 5 May 2026, according to the legalaigovernance.com carrier tracker, and Armilla offers affirmative cover that names lawyer hallucinations. Application forms have started asking whether the firm lets AI draft documents, so the policy you write now becomes the answer you give at renewal.

One policy that satisfies all seventeen

Because the opinions overlap, a single policy pitched at the strictest rule in each category clears every jurisdiction. The source of each line is in brackets.

  1. Only tools on an approved list may touch client information; the list records each tool's training and retention terms and the date they were checked. [Texas 705; California Rule 1.6 comment]
  2. Consumer accounts with default training are prohibited for client work. [Florida 24-1; ABA 512]
  3. Engagement letters disclose that AI tools may be used and describe the safeguards; matters involving self-learning tools or unusually sensitive data require written consent. [Florida 24-1; ABA 512; CCBE-style disclosure for cross-border work]
  4. No citation, quotation or factual assertion generated by AI goes into any document leaving the firm until a named lawyer has opened the source. [California 2026; Oregon 2025-205]
  5. The lawyer who signs a filing certifies personally; verification cannot be delegated to another AI tool. [Texas 705; Illinois policy]
  6. Hourly bills record actual time, including time spent prompting and checking; time saved is never billed. [Texas 705; Florida 24-1; DC 93-379]
  7. AI subscriptions are firm overhead; per-use charges are passed through at cost and only with client agreement. [California 2026; Texas 705]
  8. Fixed and value-based fees are set by reference to value and risk, and are not reduced merely because AI shortened the work. [Virginia LEO 1901]
  9. No meeting is recorded or transcribed by an AI tool without the consent of every participant. [NYC Bar 2025-6 and 2026-2]
  10. Agents that act without step-by-step prompts are supervised under the non-lawyer assistant rule, with a human gate before anything is filed or sent. [California 2026]
  11. Every lawyer completes AI training annually; attendance is logged. [Comment 8 states; NJ, FL, NC, NY CLE rules]
  12. Before every filing, the assigned lawyer checks the judge's individual rules for an AI certification requirement. [Federal standing orders; not a bar rule, but the same policy]

The law firm AI policy page turns these twelve lines into a document with a template outline, and the verification protocol implements line 4.

What to do this month

Read the opinion for your own jurisdiction once; it is short. Then read Texas 705 and the California 2026 guidance regardless of where you practise, because between them they contain every proposition the other fifteen repeat. Audit which AI accounts your lawyers actually use against line 2 above; the Texas survey suggests most firms will find consumer ChatGPT accounts doing client work. Fix the engagement letter. Module 6 of the AI Lab for Lawyers course walks through the US, UK and EU rules side by side and ends with a policy exercise built on exactly this union checklist.

Bar opinions will keep arriving, but the direction is settled: the rules are the old rules, the tool is new, and the lawyer who signs is the one who answers. The ethics question page gives the one-paragraph version for clients who ask.

Frequently asked questions

Which states have issued ethics opinions on lawyers using AI?

As of June 2026 the legalaicompliance.help tracker counts 17 jurisdictions plus the ABA. The best-known are Florida 24-1, DC 388, Kentucky E-457, Pennsylvania 2024-200, Texas 705, Virginia LEO 1901, Oregon 2025-205, Alaska 2025-1, Washington 2025-05 and Alabama 2026-01, plus New York City Bar opinions and the Illinois Supreme Court's policy. California has moved AI into its rule comments.

Do I have to get client consent before using ChatGPT on a matter?

It depends on the state and the tool. Florida Opinion 24-1 and ABA Opinion 512 call for informed consent before a self-learning tool is exposed to confidential information. Texas Opinion 705 asks for thoughtful analysis and says consent may be needed. Enterprise tools with no-training terms reduce the risk; consumer tools with default training raise it.

Can I bill a client for the time AI saved me?

No opinion allows it under hourly billing. Texas 705 and Florida 24-1 say efficiency gains inure to the client. Virginia LEO 1901 adds that a reasonable value-based or flat fee need not be cut because AI made the work faster. California's 2026 guidance treats general AI subscriptions as overhead and bars mark-ups without informed written consent.

Has any state bar banned lawyers from using generative AI?

No. Every opinion issued to date permits AI use subject to existing duties of competence, confidentiality, supervision, candour and reasonable fees. The nearest thing to a ban is at court level: two Ohio federal judges prohibit generative AI in filings except legal search engines and Microsoft products, and some districts require certification.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion

Regulation & Ethics

ABA Formal Opinion 512 Explained: What US Lawyers Must Do When Using AI

ABA Formal Opinion 512 applies six Model Rules to generative AI: competence, confidentiality, communication, candour, supervision and fees. What each duty requires in 2026.

8 min read

Regulation & Ethics

Billing Clients for AI: What You Can Charge and What You Cannot

Lawyers may bill actual time spent using and checking AI or a fixed fee agreed in advance, but not saved hours, subscriptions or marked-up AI costs. Worked numbers inside.

10 min read

Regulation & Ethics

Do You Have to Tell Clients You Use AI? Disclosure Rules and Practice

When lawyers must disclose AI use to clients under ABA 512, California, CCBE and ÖRAK rules, when consent is required, and a sample engagement-letter clause.

10 min read

Questions

Is It Ethical for Lawyers to Use AI? The Rules in the US, UK and EU

Yes, every major bar permits lawyers to use AI. It becomes unethical when output is filed unverified or client data is exposed. US, UK and EU rules in 2026.

6 min read

Use Cases

AI Meeting Notes and Transcription for Lawyers: Consent, Privilege and Setup

AI meeting notes are ethical for lawyers only with all-party consent, a no-training data path and a retention rule. Bar opinions, tools and a file-note prompt.

8 min read

Guides

Which AI Tools Train on Your Data? A Confidentiality Checklist for Lawyers

Which AI tools train on your data, plan by plan as of September 2026: ChatGPT, Claude, Gemini, Copilot, legal vendors, plus bar rules and a 10-point checklist.

11 min read