We respect attorney-client confidentiality. No tracking pixels in our emails.
California, New York, Florida, and Texas have all issued formal AI guidance for attorneys, but the rules diverge sharply. Here is what practitioners in each jurisdiction must know.
Move from this guide to practical legal AI tool shortlists and comparison pages.
Florida's Supreme Court surprised the legal technology community in early 2024 when it became the first state high court to require attorneys to certify in court filings whether they used generative AI in preparing the document. The certification requirement applied immediately to all Florida courts, created an affirmative disclosure obligation for AI-assisted legal work for the first time in any US jurisdiction, and placed Florida ahead of even California and New York in the specificity of its AI guidance. By mid-2026, more than half of US state bars have issued some form of AI-specific ethics guidance, and the requirements vary enough across jurisdictions that multi-state practitioners face a genuinely complex compliance landscape.
This guide maps the current state of AI ethics rules by jurisdiction, identifies the key differences in competence, supervision, disclosure, and fee requirements, and flags the states where rulemaking is pending and current guidance may be superseded by formal rules within the next 12 months.
The professional responsibility framework governing attorney conduct predates generative AI by decades. The ABA Model Rules of Professional Conduct, most recently amended in 2023 on unrelated topics, do not mention artificial intelligence. State bars apply existing rules to new circumstances through formal opinions, guidance letters, CLE requirements, and — in a growing number of states — formal rule amendments.
The result is a patchwork. Every state agrees that the competence obligation under Rule 1.1 applies to AI-assisted work. Every state agrees that confidentiality obligations under Rule 1.6 require careful vendor selection and data handling. Beyond those two points of consensus, the specific requirements diverge significantly, particularly around disclosure to clients and courts, supervision structures for AI-assisted work by non-attorneys, and fee reasonableness when AI dramatically reduces the time required for a task.
Practitioners operating in multiple jurisdictions — which includes most litigation attorneys, transactional attorneys at national firms, and virtually all federal practitioners — must track requirements in every jurisdiction where they are licensed or where a matter is pending. The Florida disclosure rule, for example, applies in Florida courts regardless of where the attorney is licensed.
The ABA's January 2026 Formal Opinion 512 on generative AI in legal practice is the most comprehensive national guidance to date, but it is advisory only. State bars are not bound by ABA opinions, and several have already taken positions that diverge from the ABA's recommendations.
The California State Bar's AI guidance (Formal Opinion 2023-204, supplemented by 2025 guidance) is the most detailed in the country. It addresses four core areas:
Competence: California explicitly requires attorneys to understand the specific AI tool being used, including its training data, limitations, known failure modes, and knowledge cutoff. Generalized AI literacy is insufficient — attorneys must be competent in the specific tools they deploy for client matters.
Confidentiality: Before inputting any client information into an AI tool, attorneys must assess the tool's data handling practices, determine whether inputting the data could constitute a disclosure under Rule 1.6, and obtain informed consent if disclosure is possible and consent is required. The opinion draws a meaningful distinction between tools that retain user inputs for model training (higher risk) and those that do not (lower risk but not automatically compliant).
Supervision: AI-assisted work by non-attorneys is subject to Rule 5.3. The supervising attorney must understand what the AI tool produced and cannot delegate the verification obligation.
Fees: California's guidance addresses fee reasonableness directly, noting that billing a client for the full time an AI task would have taken a human, when the AI completed the task in a fraction of that time, raises Rule 1.5 issues. The bar has not prescribed specific billing rules but has signaled that transparency with clients about AI use in fee arrangements is expected.
New York's Committee on Professional Ethics has issued multiple opinions addressing AI (Opinions 1196, 1215, and 1231). The New York framework shares California's competence and confidentiality requirements but diverges in two areas:
Disclosure to clients: New York Opinion 1215 addressed whether attorneys must disclose AI use to clients and concluded that disclosure is generally not required unless the client asks, the use involves confidentiality risks that require consent, or the fee arrangement makes the disclosure material. This is notably more permissive than what some practitioners expected.
Supervision of LLM output: New York has been particularly clear that using an LLM to draft legal documents does not satisfy the attorney's obligation to exercise independent legal judgment. The AI output must be reviewed and validated by an attorney who takes responsibility for its accuracy and appropriateness — a point that sounds obvious but has generated considerable practical guidance about what "review" must entail.
Florida's Supreme Court Administrative Order AOSC24-30 is the landmark document in this space. It requires attorneys in Florida courts to certify: whether they used AI in preparing any pleading, motion, or other document filed with the court; and if so, that a Florida Bar member reviewed the AI-generated content and certified its accuracy. This applies to generative AI tools specifically — it does not cover traditional legal research databases, spell-check, or document assembly software.
The Florida Bar has separately issued guidance addressing competence, confidentiality, and advertising rules as applied to AI-generated client communications.
The State Bar of Texas issued Formal Ethics Opinion 705 in 2024, which takes a pragmatic approach. Texas emphasizes that attorneys are responsible for the accuracy of all work product regardless of how it was produced, requires informed client consent before using AI tools with confidentiality implications, and — notably — addresses the specific scenario of AI-generated legal advice provided directly to clients through chatbots or automated intake systems. Texas requires that any client-facing AI interface clearly identify that it is not providing legal advice from an attorney unless a licensed attorney has reviewed and approved the specific output.
The ABA's opinion establishes a national baseline that most states are now referencing. Key holdings: competence under Rule 1.1 requires understanding AI tools used in a matter; supervision under Rule 5.3 applies to AI-assisted work by non-attorneys and requires meaningful review by the responsible attorney; confidentiality under Rule 1.6 requires vendor due diligence; and fees under Rule 1.5 must be reasonable, which the ABA notes may require transparency about AI use when AI dramatically affects the cost of services.
The ABA explicitly declined to recommend mandatory AI disclosure in court filings, noting that this is an area where courts rather than bar associations should set requirements.
Illinois, Washington, and Massachusetts each have active rule amendment proceedings as of mid-2026. Illinois is considering a formal rule requiring disclosure of AI use to clients. Washington is examining whether existing rules adequately address AI supervision or whether a new rule is needed. Massachusetts is conducting public comment on proposed amendments to the competence comment under Rule 1.1.
Practitioners in these states should monitor bar association announcements closely — current guidance may be superseded by formal rules within the year.
A litigation partner at a 30-attorney firm licensed in California and New York, practicing in both federal and Florida state courts, faces the following compliance matrix:
For California matters: she must conduct vendor due diligence before using any AI tool with client data, document that analysis, and ensure her billing practices reflect AI efficiencies when time savings are substantial.
For New York matters: she does not need to proactively disclose AI use to clients unless confidentiality implications require consent, but she must ensure meaningful attorney review of all AI output.
For Florida court filings: she must certify on every filing whether AI was used in its preparation. If it was, she must certify that she reviewed and verified the AI-generated content. This applies regardless of which state she is licensed in.
For federal court filings: she must check the local rules and standing orders of each specific court. Many federal district courts now have their own AI disclosure requirements that may differ from both state bar guidance and the Florida supreme court rule.
This complexity is not exceptional — it is the compliance reality for most litigation attorneys in 2026.
For staying current on state bar AI guidance and building compliant research workflows:
See also: Westlaw vs Casetext for research platform comparison.
Q: If I am licensed in New York but filing in Florida federal court, which AI disclosure rules apply to my filing?
A: Florida's Supreme Court administrative order applies to Florida state courts. For federal courts in Florida, check the specific district court's local rules and standing orders — several Florida federal courts have adopted their own AI disclosure requirements. Your New York bar license governs your professional conduct obligations; the court's local rules govern what you must certify on filings.
Q: California's guidance says I need to understand the AI tool's "limitations" — how specific does that understanding need to be?
A: Specific enough to make informed decisions about when to rely on and when to verify AI output. You do not need to understand transformer architecture, but you do need to know the tool's knowledge cutoff date, whether it is retrieval-augmented or purely generative, what categories of errors it is known to produce, and what tasks it reliably handles versus tasks where it produces unreliable output.
Q: Our firm bills hourly. If AI cuts research time from 8 hours to 2 hours, what do we bill the client?
A: This is genuinely unsettled across jurisdictions. California and the ABA have both signaled that billing the full 8 hours when AI delivered the work in 2 hours raises reasonableness concerns. The practical answer most ethics counsel give is to bill actual time plus a reasonable premium for the attorney's judgment and oversight, and to be transparent with clients about AI use in your fee arrangement.
Q: Does Florida's certification requirement apply if I only used AI for a small portion of the filing — say, generating an initial outline?
A: Yes. The Florida certification covers any use of generative AI in preparing the document, not just substantial use. If you used an AI tool at any stage of preparation, you must certify it and certify that you reviewed the AI-generated content.
Q: Are there any states that have explicitly prohibited certain uses of AI in legal practice?
A: No state has prohibited AI use categorically. Several states have issued guidance effectively prohibiting certain applications — such as using AI to generate client-facing legal advice without attorney review, or using AI tools that retain client data for model training without client consent. But the framework is restrictions on specific practices rather than categorical prohibition.
State bar compliance intersects with attorney-client-privilege obligations whenever AI systems process client communications, and with ai-hallucination risks when AI-generated research is filed with courts.
The state bar AI compliance landscape in 2026 is complex but navigable. The universal requirements — competence, confidentiality-compliant vendor selection, meaningful attorney supervision of AI output — are consistent across jurisdictions even when the specific guidance varies. The divergence is primarily in disclosure obligations, fee transparency requirements, and the formal/informal nature of the guidance.
Florida's mandatory court filing certification is the most concrete requirement currently in effect and applies to any attorney filing in Florida courts regardless of their bar licensure. Practitioners in California face the most detailed guidance on vendor due diligence. The ABA's Formal Opinion 512 provides a useful national baseline but is advisory only.
Multi-jurisdictional practitioners should build a compliance matrix for each jurisdiction where they practice that addresses the four core areas: competence documentation, confidentiality/vendor analysis, supervision structure, and fee transparency. That matrix should be reviewed whenever a jurisdiction issues new guidance — which is currently happening at a rate of several states per year.
This article reflects independent editorial analysis. LawyerAI does not accept payment for editorial coverage. Tool scores are based on methodology described in Our 5-Dimension Methodology. Last reviewed: 2026-08-23.
