Which States Require Lawyers to Disclose AI Use? A 2026 Map

The 2026 AI disclosure landscape at a glance

As of August 2026, most states have no standalone AI-disclosure mandate; disclosure to clients is triggered by materiality under ABA 512 and Rule 1.4. A few states are more specific: Florida requires disclosure when AI affects a client's costs, and California has proposed a materiality-based client-disclosure rule. Separately, many courts require disclosure in filings by standing order. Confirm your own jurisdiction, as this changes fast ^[1][2][3].

The map below is organized by the kind of rule rather than by listing fifty states, because the honest picture is that the rules cluster into a few categories and most states sit in the first one. Each entry names a verified position as of August 2026.

Client disclosure versus court disclosure

Client disclosure and court disclosure are separate duties with separate triggers. Whether you must tell your client is a materiality question under Rule 1.4, answered largely the same way in most states. Whether you must tell a court is set by that court's own rules and can require certification even where no bar rule requires client disclosure. A firm has to check both, because satisfying one does not satisfy the other.

Conflating the two questions is the most common mistake, and it runs in both directions. A lawyer who confirms the state bar has no disclosure mandate may still walk into a courtroom whose judge requires an AI certification on every filing. A lawyer focused on a court's standing order may overlook the separate duty to keep the client informed when AI materially shapes the work.

Client disclosure flows from the communication duty in Model Rule 1.4 and the materiality standard in ABA 512: tell the client when AI use significantly affects the scope, cost, or decisions in the representation. Court disclosure flows from the specific court's local rules and the individual judge's standing orders, which can require you to state whether AI was used and to certify that a human verified the output, regardless of what your bar requires.

Practically, that means two checks before a filing: the client-communication analysis under your state's version of Rule 1.4, and the specific requirements of the court and judge in front of you. Our guide on telling your client you used AI under ABA 512 covers the first; the second is a docket-by-docket question.

The states with notable positions

Florida requires client disclosure when AI use affects billing or costs, the most concrete state-level client mandate. California has proposed adding a materiality-based client-disclosure rule to Rule 1.4, which would be enforceable if adopted. Minnesota has expressly declined to require disclosure, concluding existing rules suffice. These three illustrate the spread from a specific mandate to a proposal to no mandate at all ^[2][3][4].

A few states have taken positions worth naming precisely, because they mark the edges of the range. Florida's Opinion 24-1 is the most concrete, requiring lawyers to disclose AI use where it affects the client's costs, which ties disclosure to a specific, checkable trigger rather than a general materiality judgment ^[2].

California sits at the proposal stage. Its ethics committee has advanced amendments, including a comment to Rule 1.4 addressing when a lawyer must disclose AI use to a client, triggered by significant risk or a material effect on the representation. If adopted, these would carry disciplinary weight rather than remaining advisory, which is a meaningful shift, but as of August 2026 they are proposed rather than final ^[3]. Minnesota anchors the other end: its state bar AI working group concluded that the existing Rules of Professional Conduct already govern AI use and that no separate disclosure requirement is needed ^[4].

The lesson from these three is not to memorize them but to see the range. Your state most likely sits at the materiality baseline, but it could have a specific trigger like Florida's or a pending change like California's, which is exactly why the map is a prompt to check rather than a substitute for checking.

Florida's approach deserves a second look because of how checkable it is. A materiality judgment asks a lawyer to predict whether AI use significantly affects the representation, which is a defensible but arguable call. A cost trigger asks a much simpler question: did this change what the client is billed. That is answerable from the invoice, which makes compliance easier to audit and harder to get wrong in either direction ^[2].

California's proposal is worth watching for a different reason. The distinction between advisory guidance and an enforceable rule is the distinction between a best practice and a disciplinary exposure. Advisory guidance shapes what a reasonable lawyer does; a rule comment shapes what a disciplinary body can act on. If California's amendments are adopted, the same conduct that is currently measured against guidance becomes measured against a rule ^[3].

What a court-level AI certification actually asks for

Court disclosure is where the concrete obligations live. A growing number of federal district courts require, by standing order, that filers state whether generative AI was used and certify that a human verified any AI-generated content. These orders bind you regardless of whether your state bar requires anything, and they are set judge by judge rather than statewide ^[5].

The state-bar map is the part most lawyers research and the part least likely to impose a concrete duty on a given filing. The court order is the reverse: harder to find, more likely to require something specific of you today.

These orders typically ask for two things. First, a statement of whether generative AI was used in preparing the filing. Second, a certification that a human being checked any AI-generated content, most often that the cited authorities exist and are quoted accurately. The second is the substantive one, and it maps precisely onto the verification duty that operates independently of any disclosure rule.

The important structural point is that these are per-judge instruments. Two judges in the same courthouse can have different requirements, and a lawyer who checked the district's local rules has not necessarily checked the judge's standing order. Our guide to court AI certification orders covers what these typically require and how to find the one that governs your matter.

A firm that treats the certification as a formality misreads it. Signing a statement that a human verified the AI-generated content in a filing is a representation to the court. If it turns out no one did, the problem is no longer a disclosure lapse; it is a candor problem attached to a document you signed.

What to do when you practice in more than one state

Multi-jurisdiction practice makes the patchwork a real operational problem rather than an academic one. The workable answer is to run the firm to the strictest standard it regularly encounters, rather than tracking a different disclosure posture per matter. Uniformly meeting a cost-disclosure trigger and a court certification satisfies the materiality baseline everywhere else by construction.

A firm licensed in three states with matters in several federal districts cannot realistically maintain a per-jurisdiction disclosure policy that individual lawyers apply correctly under deadline. The failure mode is predictable: the policy exists, and the person filing at 11pm applies the wrong state's version of it.

The structural way out is that these rules are nested rather than conflicting. The materiality baseline is the least demanding standard; a cost-based trigger like Florida's is more specific; a court certification is more demanding still because it requires an affirmative statement on the record. A firm that consistently meets the most demanding standard it regularly faces has satisfied the others without tracking them individually.

In practice that means three defaults. Tell clients about AI use in the engagement letter rather than deciding materiality matter by matter. Verify every cited authority regardless of whether the court in front of you demands a certification. Keep the record of that verification whether or not anyone asks for it.

None of those defaults requires knowing which state you are in, which is the point. They are cheaper to run than a jurisdictional decision tree and they fail safe. The residual work is narrow: check the specific judge's standing order before each filing, because that is the one requirement a firm-wide default cannot anticipate.

This does not replace confirming your obligations in each jurisdiction where you are licensed, and it is not legal advice about any state's rules. It is an operational posture that reduces how often the jurisdictional question has to be answered correctly under pressure.

The ABA Opinion 512 baseline every state builds on

ABA Formal Opinion 512 is the national baseline, and it does not impose blanket AI disclosure. It frames disclosure as a function of the existing duties: communicate with the client under Rule 1.4 when AI materially affects the representation, protect confidentiality under Rule 1.6, and maintain competence under Rule 1.1. Most state positions are variations on this materiality approach rather than departures from it ^[1].

Understanding 512 explains why the map looks the way it does. Rather than creating a new AI-specific disclosure rule, the ABA opinion reads the existing Model Rules onto AI use, and most states have followed that lead instead of legislating a standalone mandate.

Under 512, disclosure to a client is required when it is material to the representation, for example when AI use significantly affects cost, strategy, or the client's decisions, which is a judgment call grounded in Rule 1.4 rather than a checkbox. Confidentiality under Rule 1.6 governs what client data may go into a tool, and competence under Rule 1.1 requires understanding the tools well enough to use them responsibly. Because most states adopt this framing, the practical rule in most of the country is not "always disclose" but "disclose when it matters."

That baseline is stable even as specific states move, which is why it is the right anchor. A firm that internalizes the materiality standard is prepared for most jurisdictions and only needs to layer on the specific state and court variations where they exist.

How to check your state and keep up

Check three sources for your jurisdiction: your state bar's AI ethics opinion or guidance for client-disclosure rules, the specific court's local rules and the judge's standing orders for court disclosure, and any pending rule changes at your state bar. Because this area changes quickly, recheck before relying on a prior answer, especially in states like California with proposals in progress. Treat any map, including this one, as dated.

The durable skill is not knowing every state's rule today but knowing where to look and how often. Three sources answer the question for any jurisdiction.

First, your state bar's AI ethics opinion or guidance, which sets the client-disclosure standard and tells you whether your state has moved beyond the materiality baseline. Our guide on finding and reading your state's AI guidance walks through this. Second, the local rules of the specific court and the individual judge's standing orders, which govern court disclosure and are often stricter than any bar rule; a growing number of courts require an AI certification in filings, so check the docket you are filing in. Third, pending changes, because proposals like California's can become enforceable rules on a timeline that outruns a static guide.

Set a habit of rechecking rather than relying on a prior answer. This guide reflects the landscape as of August 2026, and in an area moving this fast, a dated map is a prompt to verify, not a conclusion to trust.

This map reflects the AI disclosure landscape as of August 2026 and is educational, not legal advice. State bar rules and court orders on AI disclosure change frequently; confirm the current requirement with your state bar or a licensed attorney in your jurisdiction before relying on it. The California rule described here was proposed, not final, as of August 2026.