No state open meetings act currently requires your minutes to carry a label saying AI helped draft them, and the AI disclosure statutes usually cited are scoped to systems that interact with the public, not to a drafting tool your clerk runs after the meeting. That is not a reason to say nothing. It is a reason to disclose in the right place: adopt a written AI-use policy, which becomes a public record in its own right and answers the question a requester will actually ask. Keep the generic “AI-generated, may contain errors” disclaimer off the minutes themselves, because the legal force of minutes comes from the clerk's certification and the body's approval vote, and a disclaimer of accuracy undercuts both.
The question comes up within about a month of a board adopting any kind of automated drafting, and it usually arrives from a cautious member or a cautious attorney. Shouldn't the minutes say that a computer helped write them?
It is a good instinct pointed at the wrong target. The instinct is that the public deserves to know how its record was made, which is correct. The target is wrong because a footer on the document is the least informative and most legally awkward way to tell them.
What follows is the actual legal landscape as of September 2026, what a well-designed disclosure rule looks like when a profession has genuinely worked the problem through, and the policy language a public body should adopt instead.
The laws people cite mostly do not say what people think they say
Start with the honest answer, because a lot of what circulates on this topic is secondhand and wrong: there is no state open meetings act that requires a set of minutes to disclose on its face that generative AI assisted in producing it. If someone tells you otherwise, ask for the section number and read the section. More than one widely repeated claim in this area turns out, on inspection, to cite a statute about something else entirely.
The statute most often invoked is Texas. The Texas Responsible Artificial Intelligence Governance Act, House Bill 149, took effect January 1, 2026, and it does impose a genuine disclosure duty on government. Texas Business and Commerce Code § 552.051 provides that “a governmental agency that makes available an artificial intelligence system intended to interact with consumers shall disclose to each consumer, before or at the time of interaction, that the consumer is interacting with an artificial intelligence system.” The disclosure “must be clear and conspicuous,” “must be written in plain language,” and “may not use a dark pattern.”
Read the scope clause, because it decides everything. The duty attaches to a system “intended to interact with consumers,” and it is discharged “before or at the time of interaction.” A transcription and drafting tool that a clerk runs against a recording after the meeting has ended never interacts with a member of the public at all. There is no interaction to disclose at. If you put an AI assistant on your public meeting portal so residents can ask it questions, that is squarely within the statute and you should treat it that way. Your minutes drafting workflow is not.
The federal records authority has been just as careful. On August 21, 2026, the National Archives issued AC 11.2026, Guidance on Applying the Federal Records Act to Artificial Intelligence Materials. Its central holding is that agencies' “use of AI platforms does not, in and of itself, result in the creation of federal records,” and that “subsequent use of AI-generated materials in agency business may result in the creation of federal records.” Whether something is a record “depends on the circumstances surrounding the creation, maintenance, and use of the materials, such as whether the agency relies on it in decision-making, uses it to conduct official business, circulates the material to others, or incorporates it into an agency system.”
Then, in a sentence worth sitting with, NARA draws a boundary around its own guidance: the memo “addresses records management requirements under the FRA only and does not establish policy related to AI governance, e-discovery, privacy, security, or ethical use.” The agency that governs the federal record looked directly at the disclosure question and declined to answer it, because it is not a records question. It is a governance question, and governance is set by policy.
Two caveats on that source, stated plainly. NARA binds federal agencies, not your city council or school board, so this is persuasive reasoning rather than law that applies to you. And the guidance is three weeks old at the time of writing, so state archives have not yet had a chance to respond to it.
What a disclosure rule looks like when someone designs it properly
One profession has already fought this out in public and landed somewhere instructive: the courts. After a run of filings containing fabricated citations, federal judges began requiring lawyers to say whether AI wrote any part of a brief. What began as individual standing orders has hardened into a rule of general application.
Local Civil Rule 7.2(f) of the United States District Court for the Northern District of Texas, in the edition effective September 1, 2026, is worth quoting in full because the structure is the lesson:
(1) A brief prepared using generative artificial intelligence must disclose this fact on the first page under the heading “Use of Generative Artificial Intelligence.” If the presiding judge so directs, the party filing the brief must disclose the specific parts prepared using generative artificial intelligence.
(2) “Generative Artificial Intelligence” means a computer tool (whether referred to as Generative Artificial Intelligence or by another name) that is capable of generating new content (such as images and text) in response to a submitted prompt (such as a query) by learning from a large reference database of examples.
(3) A party who files a brief that does not contain the disclosure required by subsection (f)(1) of this rule certifies that no part of the brief was prepared using generative artificial intelligence.
Three properties make this a real rule rather than a gesture, and all three are missing from the footer most boards are contemplating.
It attaches to a document, not to an organization. The disclosure is about this brief, filed on this day. It is not a standing announcement that the firm owns software.
It defines the term. Subsection (f)(2) exists so that nobody can litigate whether their tool counts. Note how broad it is: anything that generates new content from a prompt by learning from a large reference database, “whether referred to as Generative Artificial Intelligence or by another name.” A body writing its own policy should be equally specific, because otherwise the first genuine question, does the transcription engine count, has no answer.
Most importantly, silence means something. Subsection (f)(3) is the engineering. Filing without the disclosure is itself an affirmative certification that no AI was used. That single clause converts the rule from an honor system into a binary with consequences at both ends, and it is why the rule has teeth that a voluntary disclaimer never will.
Notice also what the rule does not do. It does not prohibit using the tool, and it does not ask the filer to warn the court that the brief may be inaccurate. Rule 11 already makes the signing attorney answerable for every word. The disclosure sits alongside responsibility; it does not dilute it.
Why a blanket disclaimer makes your record weaker
Here is the part that gets missed, and it is the reason to be firm about this rather than agnostic.
Minutes are not just a document that describes a meeting. They are an instrument with evidentiary force, and that force is attached to the clerk. Illinois states the mechanism as directly as any jurisdiction. Under 65 ILCS 5/3.1-35-90, the municipal clerk “shall attend all meetings of the corporate authorities including executive sessions and keep a full record of their proceedings in the journal,” and subsection (c) provides that “copies of all papers duly filed in the clerk's office and transcripts from the journals and other records and files of the clerk's office, certified by the clerk under the corporate seal, shall be evidence in all courts in like manner as if the originals were produced.”
Read that last clause slowly. The clerk's certification is what lets a copy of the minutes walk into a courtroom and stand in for the original. It is a load-bearing element of the document, not a formality at the bottom of the page.
Now attach the sentence boards keep proposing: These minutes were generated by artificial intelligence and may contain errors.
That is not a disclosure. It is a disclaimer of accuracy, published by the body, on the face of the body's own official record, under the signature of the officer whose certification is the thing giving the document force. You have taken an instrument whose entire value is that it is reliable and printed on it a statement that it may not be. The first time a decision recorded in those minutes is challenged, the challenger does not need to attack your record. You have written their sentence for them and put your seal under it.
Vendors bear some responsibility for the confusion. An “AI may make mistakes” notice is standard in consumer software, where it is a liability shield for the vendor. It does not belong on a governmental record, and copying it across is how a sensible product convention becomes a bad legal habit.
Your record already has a disclosure mechanism, and it is stronger
The public's real question is not whether a machine was in the loop. It is: who is answerable for this being right, and what did they check it against?
Minutes already answer that question, twice. The clerk certifies the record. The body adopts it by vote, and that vote is an attestation by the people who were in the room that the document is an accurate account of what happened there. Those two acts are the accountability structure, and neither is affected by what software produced the first draft. A clerk who verified a machine draft against the recording has done the same verification job as a clerk who typed it from shorthand, and is making the same statement by signing.
A footer saying AI was involved does not add to that. It substitutes a fact about tooling for a statement about responsibility, and it invites the reader to discount a document that two separate accountability mechanisms have already stood behind.
What actually belongs on the record: an adopted policy
The disclosure that does the work is a written AI-use policy, adopted by the body in an open meeting. It is a public record from the moment it is adopted, and unlike a footer it answers a records request on its own. NARA reaches the same conclusion for federal agencies, recommending that they adopt formal AI policies developed with legal, IT, and other stakeholders rather than relying on ad hoc practice.
A policy that covers minutes should answer seven questions specifically enough that a member of the public reading it knows what happened to their meeting.
- Which tools, and for which steps. Name them. Distinguish transcription from summarization from drafting, because they carry very different risk and a blanket “we use AI” tells nobody anything.
- Who verifies, and against what. This is the most important line in the policy. The answer should name a source, and for minutes the source is the recording. Verification against nothing is review theater.
- What is never delegated. Motion language, the vote and how each member voted, names, and anything arising from closed session. These are the fields a challenge will be about, and the policy should say a human writes or confirms them every time.
- What happens to the artifacts. The transcript, the prompts, the intermediate drafts. Which retention schedule they fall under and who is responsible for them.
- Where the data goes. Which vendor, held how long, and whether your meeting audio is used to train anyone's model. Executive session recordings make this question sharper than most bodies realize.
- How the policy is adopted and reviewed. By vote, in public, with a date for revisiting it. A policy sitting in a staff folder is not a disclosure to anybody.
- What the clerk is certifying. Spell out that the certification is a statement about verification, so the meaning of the signature does not quietly drift when the drafting method changes.
Adopt that and you have disclosed far more than any footer conveys, in a document designed to be read by the people who care, without writing a word of doubt onto the record itself.
The retention half nobody plans for
One consequence of the policy deserves its own paragraph, because it is where bodies get caught.
The transcript your minutes were drafted from is very probably a record. The Texas State Library and Archives Commission put the general principle plainly in its guidance on classifying AI-generated records: “AI-generated records should be classified the same as human-generated records,” and “the final product of a record should be classified based on the administrative purpose and function the record serves.” Retention follows function, not origin. The commission is equally direct that staff remain accountable for records created with AI assistance on the entity's behalf.
Apply NARA's factors and you reach the same place. Did you rely on the transcript in conducting official business? Did you circulate it? Did it go into a system? For a transcript that the official minutes were drafted from, those answers are yes, and the instinct to delete it once the minutes are approved is exactly backwards. We have written before about how AI transcripts sit inside the public record; the policy point here is narrower and more practical. Name the schedule in writing, in advance, so that nobody is improvising a retention decision under the pressure of a request that has already arrived.
If you do put a line on the document, write it like this
Some bodies will want something on the page anyway, and there are decent reasons: a community with low trust in its government, a board that has taken public criticism, a clerk who would simply rather be explicit. That is a legitimate policy choice, and it can be made well.
The rule is to describe the process and end on the human. Something like:
These minutes were prepared by the Clerk from the official audio recording of the meeting, with the assistance of automated transcription and drafting software, and were reviewed by the Clerk against that recording before publication.
Every clause earns its place. It names the source of truth, which is the recording. It is honest about the software without overstating its role. It names the officer. It ends on verification rather than on doubt, and it contains no statement that the document may be inaccurate. Put it in the front matter with the date, time, and attendance, where process notes belong, rather than as a disclaimer at the foot of the page.
What it does not do is warn the reader off the document. That distinction is the whole of the argument.
The through-line
Disclosure is not a label. It is accountability, and the two get confused because a label is so much easier to produce.
The courts got this right the first time. Rule 7.2(f) does not ban the tool and does not ask anyone to caveat their own filing. It asks a specific question about a specific document, defines its terms so the question can be answered, and makes silence an answer too, all while leaving the signing attorney fully responsible under Rule 11. Responsibility was never in question; the rule just made the facts visible.
Minutes need the same treatment, and mostly already have it. The clerk certifies. The body adopts. What is missing in most jurisdictions is not a footer, it is a written policy saying which tools are used, who checks them, against what, and what happens to everything the process leaves behind.
So put the policy on the record, keep the certification on the document, and leave the disclaimer off. The public deserves to know how its record was made. It does not deserve a record that announces it might be wrong.
Frequently asked questions
Do we have to disclose that AI helped write our meeting minutes?
As of September 2026, no state open meetings act requires minutes to carry a label saying artificial intelligence assisted in drafting them. The AI disclosure statutes usually cited are scoped differently. Texas, for example, requires a governmental agency that makes an AI system available to interact with consumers to disclose that fact before or at the time of interaction, which reaches a public-facing chatbot rather than a drafting tool a clerk runs on a recording after the meeting. Before accepting any claim that your state mandates disclosure in the minutes themselves, ask for the section number and read it.
Should we add an “AI-generated, may contain errors” disclaimer to our minutes?
No. That sentence is not a disclosure, it is a disclaimer of accuracy, and it is being attached to a document whose entire legal value rests on being accurate. Minutes carry evidentiary weight because a clerk certifies them and a body adopts them. A line telling readers the document may be wrong undercuts both, and it hands a challenger your own record to quote against you. If you want a note on the document, describe the process and name the human who verified it instead.
Do we have to keep the AI transcript and the automated draft?
Usually yes, and you should assume so until your records officer tells you otherwise. The governing principle in state archives guidance is that records are classified by the administrative purpose and function they serve, not by whether a human or a machine produced them. Federal guidance points the same way, treating material as a record based on whether the agency relies on it in decision making, uses it to conduct official business, circulates it to others, or incorporates it into an agency system. A transcript you drafted the official minutes from meets most of those tests. Deleting it to avoid a records request is the worst available option.
Does the Texas AI law apply to our minutes software?
Probably not in the way it is often described. The Texas Responsible Artificial Intelligence Governance Act, effective January 1, 2026, requires a governmental agency that makes available an artificial intelligence system intended to interact with consumers to disclose to each consumer, before or at the time of interaction, that the consumer is interacting with an artificial intelligence system. The disclosure must be clear and conspicuous, written in plain language, and may not use a dark pattern. Minutes drafting software is used by staff rather than by the public, so the interaction the statute regulates does not occur. A public-facing meeting search assistant on your portal is a different question.
What should a public body's AI-use policy for minutes actually cover?
Seven things: which tools are approved and for which steps; who verifies the output and against what source; which parts of the record are never produced without direct human authorship, meaning motions, votes, names, and anything from closed session; how long the transcript, prompts, and drafts are retained and under which schedule; where the data goes and whether the vendor trains on it; how the policy itself is adopted, which should be by vote in an open meeting; and what the clerk is certifying when they sign. Written that way, the policy answers a records request on its own.
Does using AI mean the minutes are no longer the clerk's work?
No, and state records guidance is explicit that the human remains accountable regardless of the tool. Staff bear responsibility for records created with AI assistance on behalf of their government entity, exactly as they do for a document typed up from handwritten notes. The clerk's certification is a statement about verification, not a claim of authorship over every keystroke, which is why it survives the introduction of a drafting tool and why it is the disclosure that matters most.
Sources: Tex. H.B. 149, 89th Leg., R.S. (2025) (Texas Responsible Artificial Intelligence Governance Act), enacting Tex. Bus. & Com. Code § 552.051, eff. Jan. 1, 2026 · N.D. Tex. Local Civil Rule 7.2(f), Disclosure of Use of Generative Artificial Intelligence (Local Civil Rules effective September 1, 2026) · National Archives and Records Administration, AC 11.2026, Guidance on Applying the Federal Records Act to Artificial Intelligence Materials (Aug. 21, 2026) · 65 ILCS 5/3.1-35-90 (Illinois Municipal Code, municipal clerk duties) · Texas State Library and Archives Commission, Classifying AI-Generated Records (June 10, 2024). NARA guidance governs federal agencies and is cited here as reasoning, not as law binding local government. Statutes are cited as illustrations of a cross-cutting principle rather than as a survey, and AI legislation is moving quickly, so check your own state's current law and your attorney's advice. This article is general information, not legal advice.
Ryan Wilson is the founder and CEO of Govably, which builds AI-assisted agenda and minutes software for city, county, and school-district clerks.