“How do you underwrite it if one state requires you to do something and another state doesn’t?” That was Al Roberts, head of professional liability at Berkley Select, talking to the Daily Record in August 2026. It is the underwriter’s problem, and yours if you hold a licence in more than one state, because the state bar AI ethics opinions issued since 2023 agree on principles and diverge on the details that bite: consent before client information touches a tool, whether you must tell the client, and what you may bill.
Pennsylvania wants lawyers “transparent with clients, colleagues, and the courts”; New Jersey finds no routine duty. Florida “recommends” consent; North Carolina requires it only for delegated substantive tasks. What follows is a map: one table, a note on each state that adds something, the technology-CLE mandates, a strictness heat-map and the direction of travel, with dates as of 13 September 2026.
How to read the table
Six duties recur in every opinion: competence (Rule 1.1 and Comment 8), confidentiality (1.6), communication and consent (1.4), reasonable fees (1.5), candour (3.1 and 3.3) and supervision (5.1 and 5.3). ABA Formal Opinion 512 of 29 July 2024 is the template most states quote, so the table records only what each state adds. These are advisory opinions; the binding law is your state’s rules. And “state” is loose: two of the most useful opinions come from New York City’s bar association, and the Illinois entry is a Supreme Court policy.
| State | Instrument (date) | Billing rule | The distinctive addition |
|---|---|---|---|
| California | COPRAC guidance (Nov 2023; rewritten May 2026) | Actual time; subscriptions are overhead; pass-through at cost | Guidance on agentic AI; proposed Rule 3.3 verification duty |
| Florida | Ethics Opinion 24-1 (19 Jan 2024) | No inflated time; no charge for “minimal competence”; no prorating of subscriptions | Consent “recommended”; advertising claims “objectively verifiable”; chatbots must identify themselves |
| New York City | Formal Opinion 2024-5 (7 Aug 2024) | Follows ABA 512 | Open versus closed systems; no client data into open systems without consent |
| New Jersey | Supreme Court Committee guidelines (Jan 2024) | Existing rules | Disclose only “if a client asks” or the client cannot decide without knowing |
| Pennsylvania / Philadelphia | Joint Formal Opinion 2024-200 (May 2024) | AI expenses “appropriately disclosed” | Transparency by default; proficiency “to the same extent” as traditional methods |
| Kentucky | KBA E-457 (March 2024) | Reduce fees when AI reduces time | Disclosure required when costs are billed or a court requires |
| District of Columbia | Ethics Opinion 388 (April 2024) | “Only actual time spent” | Quotes NPR’s “omniscient, eager-to-please intern who sometimes lies to you” |
| Texas | Opinion 705 (Feb 2025) | May not “charge hourly fees for the time that was ‘saved’” | “Reasonable and current understanding”; “cannot blindly rely” |
| North Carolina | 2024 FEO 1 (1 Nov 2024) | Three hours to one: bill one; flat fees allowed | “Advanced informed consent” for delegated substantive work; vendors are nonlawyer assistants |
| Virginia | State Bar guidance (2024) | “May not bill for time saved” | “No per se requirement to inform a client” |
| Illinois | Supreme Court AI Policy (Jan 2025) | Not addressed | Use “should not be discouraged”; no pleading disclosure |
California: an agentic rewrite and a bill on the Governor’s desk
California moved first and keeps moving. COPRAC’s guidance appeared on 16 November 2023; on 14 May 2026 the Board of Trustees approved a rewrite that, at the California Supreme Court’s request, addresses agentic AI: “Lawyers must not permit AI systems to autonomously file documents, communicate with the court, or make representations on the lawyer’s behalf.” “The greater the level of system autonomy, the greater the lawyer’s obligation to implement oversight mechanisms sufficient to ensure that professional judgment remains with the lawyer.” And: “The lawyer’s duty of candor to the tribunal cannot be delegated to AI.”
It is also the most precise state on money: charge “for actual time spent (e.g., crafting or refining generative AI inputs and prompts, or reviewing and editing generative AI outputs)”; subscriptions are overhead “similar to library maintenance”; matter-specific costs pass through with “no markup or profit element”. Competence “includes periodic reassessment of the system’s capabilities and risks”.
Proposed amendments to Rules 1.1, 1.4, 1.6, 3.3, 5.1 and 5.3 would write the guidance into the rules, with Rule 3.3 requiring lawyers to “verify the accuracy and existence of cited authorities”; a revised proposal was out for comment until 6 August 2026. SB 574 (Umberg), which would bar delegating the practice of law to generative AI, passed the legislature on 31 August 2026 and went to the Governor on 9 September; as of 13 September 2026 it is not law.
Florida 24-1: consent, fees and advertising
Ethics Opinion 24-1 (19 January 2024): “It is recommended that a lawyer obtain the affected client’s informed consent prior to utilizing a third-party generative AI program if the utilization would involve the disclosure of any confidential information.” Note the verb. And a lawyer may never delegate “any act that could constitute the practice of law such as the negotiation of claims”.
Efficiency “must not result in falsely inflated claims of time”; do not charge “for the time spent developing minimal competence”; and where a tool’s per-matter cost cannot be determined, “the lawyer may not ethically prorate the periodic charges of the generative AI”. Advertising: lawyers “cannot claim their generative AI is superior to those used by other lawyers or law firms unless the lawyer’s claims are objectively verifiable”.
New York City: open systems, closed systems and the note-taker opinion
Formal Opinion 2024-5 (7 August 2024) gave the profession its most usable vocabulary: “open” systems share inputs with third parties, “closed” systems keep them inside the firm. “Without client consent, a lawyer must not input confidential client information into any Generative AI system that will share the inputted confidential information with third parties.” Even with consent, “avoid entering details that can be used to identify the client”; and terms of use “can change frequently”, so the duty to understand them “is continuing”.
Formal Opinion 2025-6 (22 December 2025) is the first bar opinion on AI note-takers: consent before recording, and a warning that transcripts “could effectively formalize everything that was said, including informal legal advice that an attorney might have offered to the client ‘off the cuff’”. See the note-taker guide.
New Jersey and Pennsylvania: the disclosure split
Ask “do I have to tell my client?” on either side of the Delaware and you get different answers. New Jersey’s preliminary guidelines (25 January 2024) require disclosure only “if a client asks if the lawyer is using AI, or if the client cannot make an informed decision about the representation without knowing that the lawyer is using AI”. Pennsylvania and Philadelphia’s Joint Formal Opinion 2024-200 (22 May 2024) says lawyers “should be transparent with clients, colleagues, and the courts about the use of AI tools” and “must be proficient in using technological tools to the same extent they are in employing traditional methods”. The decision tree is in Do lawyers have to disclose AI use?
North Carolina 2024 FEO 1: three hours to one
On disclosure: “Generally, a lawyer need not inform her client that she is using an AI tool to complete ordinary tasks”, but delegating substantive tasks needs “the client’s advanced informed consent”. What to do with the two hours you no longer bill is in AI and the billable hour.
Kentucky, DC, Texas, Virginia, Illinois and the rest
Kentucky’s KBA E-457 (March 2024) answers seven questions: competence is mandatory, fees come down when AI cuts the time, and disclosure is required when AI costs are billed or a court requires it. DC’s Ethics Opinion 388 (April 2024) borrows NPR’s description of a chatbot as “an omniscient, eager-to-please intern who sometimes lies to you”; hourly billers charge “only actual time spent”.
Texas Opinion 705 (February 2025) requires “a reasonable and current understanding of the technology”, says “lawyers cannot blindly rely upon or use answers given by generative AI tools”, and bars charging “hourly fees for the time that was ‘saved’”. The State Bar’s 2026 survey of 1,553 lawyers shows why: AI use rose from 30% in 2024 to 62%, half report no firm policy, and 67% support required CLE on the ethical use of AI.
Virginia’s guidance (press-dated August 2024) is blunt on both contested points: “may not bill for time saved by using generative AI” and “no per se requirement to inform a client”. The Illinois Supreme Court’s Policy on Artificial Intelligence (effective 1 January 2025) says AI use “may be expected, should not be discouraged, and is authorized provided it complies with legal and ethical standards. Disclosure of AI use should not be required in a pleading.” Permissive has a limit: in July 2026 the Illinois Appellate Court fined a lawyer $15,000 for ten false citations and quotations and held that “no matter how much one pays for ‘premier’ or ‘corporate’ versions of AI products, it does not negate an attorney’s obligation to verify all citations of authority” (Scott v. Illinois Human Rights Commission).
Alaska issued Ethics Opinion 2025-1 in April 2025, Mississippi a scenario-based guide, and the NYSBA’s 85-page task-force report (April 2024) warns against “techno-solutionism”; one tracker counts “more than twenty-five” state bars with guidance. If your state is missing above, check before assuming silence.
Using only [state bar ethics opinions page URL], [state supreme court news page URL] and the attached PDF of [opinion number] dated [date]: has [state] issued, amended or withdrawn guidance on lawyers' use of generative AI since [date]? For each item give title, issuing body, date, URL and a two-sentence summary of what it adds on (a) client consent, (b) disclosure to clients, (c) billing, (d) verification of citations, (e) agentic tools. Where you find nothing, write "NOTHING FOUND ON THE SOURCES PROVIDED". Do not draw on your own knowledge; quote the source or say nothing.Which states require technology CLE
Comment 8 to Model Rule 1.1 (2012) has been adopted by 40 states plus DC and Puerto Rico; only three states put hours behind it.
| State | Requirement | Since |
|---|---|---|
| Florida | 3 hours of technology CLE per 3-year cycle | 1 January 2017 (first in the US) |
| North Carolina | 1 hour a year on information technology or cybersecurity | 2019 |
| California | At least 1 hour on technology in the practice of law | Current requirement |
None mandates AI content, though AI programmes qualify; courts have begun ordering AI-ethics CLE as a sanction. The credit picture is in the technology CLE guide.
The strictness heat-map
My reading, not the bars’: “strictest” means a specific, testable obligation beyond ABA 512.
| Consent before client data enters a tool | Telling the client | Billing | Verification and autonomy | |
|---|---|---|---|---|
| Strictest | New York City; North Carolina (delegated substantive work); Kentucky | Pennsylvania | Florida; California; Texas and Virginia | California (no autonomous filing; proposed Rule 3.3) |
| Middle | California; Florida (recommended) | ABA 512, DC, Kentucky | North Carolina; DC | Texas; Florida |
| Most open | New Jersey | New Jersey, North Carolina, Virginia | Illinois (silent) | Illinois (no pleading disclosure) |
A multi-state firm should write to the strictest column it practises in: New York City on consent, Pennsylvania on disclosure, Florida and California on billing, California on autonomy. That policy satisfies everyone else on the list.
Review our generative AI policy <policy>...</policy> against each attached state opinion <opinions>...</opinions> for [states where our lawyers are licensed]. Build a table with one row per obligation (client consent before inputting client information; disclosure to clients; billing for actual time; subscriptions and pass-through costs; verification of citations; supervision of vendors; agentic tools) and one column per state. In each cell quote the operative sentence, then mark the policy MEETS / GAP / SILENT. For each row name the strictest state and draft the one sentence that would satisfy it. Use only the attached texts; where an opinion is silent, write SILENT.The trend: from “may use” to “must supervise”
Read the opinions in date order and the verb changes: Florida (January 2024) permits use where the lawyer “can reasonably guarantee compliance”, Illinois (January 2025) says use “should not be discouraged”, California (May 2026) says what the system must never do on its own. The subject has moved from the lawyer’s use to the tool’s autonomy, and from policy to proof.
Two cases explain the move. In Johnson v. Dunn (N.D. Ala., 23 July 2025) Butler Snow had a written AI policy since June 2023; three partners were still publicly reprimanded, disqualified and referred to the Alabama bar over ChatGPT citations. Judge Manasco: “They benefitted from repeated warnings, internal controls, and firm policies about the dangers of AI misuse… And yet here we are.” In Prososki v. Regan (Nebraska Supreme Court, 20 March 2026) the court struck the brief and referred the lawyer to the Counsel for Discipline (“It does not matter if [generative AI] told you so”); in April 2026 local press reported a temporary suspension “until further notice”. The first suspension tied to AI was People v. Crabill (Colorado, November 2023); the rest are in the sanctions timeline.
Draft an engagement-letter clause on our use of generative AI for a firm with lawyers licensed in [Florida, New York and California], satisfying the strictest of the three on each point; ABA Formal Opinion 512 says boilerplate consent "is not sufficient". Cover: the tools we use (enterprise tiers with no-training terms; no consumer tools for client information); what client information may be entered and the client's right to instruct otherwise; the specific risks and benefits; our human-review commitment; billing (actual time only; no charge for learning tools; subscriptions as overhead; metered matter costs at cost with prior written consent); and how consent can be withdrawn. Plain English, under 350 words, then a one-line note naming which state drove each sentence.Where to go next: the jurisdiction-by-jurisdiction guide covers the UK, EU and DACH rules; court standing orders explains the judge-by-judge certification layer no state opinion touches; the rest of the ethics and regulation hub is here and the prompts are in the prompt library. If your firm spans several of these states, the way through is one working practice built to the strictest of them, which is what participants build on their own documents in AI Lab for Lawyers.
Frequently asked questions
Which states have issued AI ethics opinions for lawyers?
As of September 2026 the primary documents are California's COPRAC guidance (2023, rewritten 2026), Florida 24-1, New York City Bar 2024-5 and 2025-6, the NYSBA task-force report, New Jersey's preliminary guidelines, Pennsylvania/Philadelphia 2024-200, Kentucky E-457, DC 388, Texas 705, North Carolina 2024 FEO 1, Virginia's guidance, Illinois's Supreme Court policy, Alaska 2025-1 and Mississippi's practical guide. One tracker counts more than twenty-five state bars with some form of guidance.
What does California say about lawyers using AI?
California's Practical Guidance, rewritten and approved on 14 May 2026, addresses agentic AI, which no other opinion in this map does: lawyers must not let systems autonomously file documents or make representations, and the duty of candour cannot be delegated. It treats AI subscriptions as overhead and allows matter costs to pass through only at cost. Proposed amendments to Rules 1.1, 1.4, 1.6, 3.3, 5.1 and 5.3 went out for a second comment period ending 6 August 2026; SB 574 awaits the Governor.
Does Florida require client consent for AI?
Florida Ethics Opinion 24-1 (19 January 2024) recommends, rather than mandates, obtaining the client's informed consent before using a third-party generative AI tool where confidential information would be disclosed, and says no consent is needed where nothing is shared with a third party. Lawyers must know whether a tool is self-learning. The opinion also bars inflated time claims, charging for developing minimal competence, prorating subscriptions, and unverifiable claims that your AI is superior.
Which states require technology CLE?
Three. Florida was first, with three hours of technology CLE per three-year cycle from 1 January 2017. North Carolina requires one hour a year on information technology or cybersecurity, from 2019. California requires at least one hour on technology in the practice of law. None mandates AI-specific content, though AI ethics programmes qualify. Separately, the Comment 8 duty of technology competence has been adopted by 40 states plus DC and Puerto Rico.
Are state AI opinions binding on lawyers?
No. They are advisory interpretations of your state's Rules of Professional Conduct, which are binding; Illinois's entry is a Supreme Court policy and California is proposing to write AI duties into the rules themselves. Disciplinary bodies read the rules through these opinions, so departing from them is risky even where they are not law. Judge Manasco's point in Johnson v. Dunn applies: a policy nobody follows protects nobody.