This post is the blog edition of the first Legal AI Governance newsletter. It covers April 1 through September 30, 2026. The tracker was brought current in late September and early October, adding 286 case entries and 131 court-order entries, so it runs longer than a normal issue will.
Each item was checked against the primary document: the rule, the order, the opinion. Each item links that document on first reference, and the tracker entry that records it. Statements about the status of a rule or proceeding are as of September 30, 2026. The figures count tracker entries as coded on 2026-10-01, not every order a court issued. This post is informational and is not legal advice; firms weighing any of this should consult a licensed attorney in their state.
The shift: from disclosing AI use to certifying citations
Florida made the clearest move. On 2026-05-28 the Supreme Court of Florida amended Rule of General Practice and Judicial Administration 2.515(d)(2) on its own motion (In re: Amendments to Florida Rule of General Practice and Judicial Administration 2.515, No. SC2026-0673 (Fla. May 28, 2026)). Since 2026-06-15, each signer of a document filed in a Florida court represents that “the legal authorities identified exist and are accurately cited.” The rule text does not mention artificial intelligence, and nothing in it asks whether AI was used.
A companion administrative order, In re: Representations by Signers of Filings, No. AOSC26-12 (Fla. May 28, 2026), went further. Florida courts “may not impose” AI disclosure or certification requirements, “whether through local administrative orders, court policies, judicial practices and procedures, or other means.” Circuit orders in the 11th, 15th, 17th, 18th and 19th Circuits were rescinded or vacated, all effective by 2026-06-15. The amendment took effect before public comment, which closed 2026-08-11. As of September 30, 2026, the court had issued no further order on it. (Tracker entry)
Other state courts wrote the same idea into their own rules:
- New York. 22 NYCRR Part 161 (tracker) took effect 2026-06-01 across the Unified Court System. Its statewide policy says attorneys and parties “should not be required” to disclose AI use. Each court may adopt its own AI part rule and is encouraged to use a Model Rule, under which signing a paper certifies a careful, independent review for fabricated authority.
- Connecticut. Practice Book Section 4-9 (tracker), effective 2026-06-23, requires anyone who uses generative AI in a Superior Court filing to “independently verify all citations, legal authorities or evidence” it produced. The appellate rules (Practice Book Sections 60-4, 62-6 and 85-2) (tracker) adopted the same duty effective 2026-07-14.
- Indiana. Amended Trial Rule 11 (Order No. 26S-MS-8) (tracker), effective 2026-08-01, makes an attorney’s signature certify “that any cited legal authority is authentic,” whether or not AI helped prepare the filing.
- Texas (proposed). On 2026-06-26 the Supreme Court of Texas proposed a new Rule of Civil Procedure 13(c)(1) (Misc. Docket No. 26-9054; tracker). Signing a filing would represent that the signer verified it, “including that the legal citations, quotations, and holdings are not fabricated.” Comments closed 2026-09-01. The court had said it expected the rule to take effect 2026-10-01, but it had posted no final order as of September 30, 2026.
A separate post, Statewide AI Court Rules in 2026: Verification, Not Disclosure, goes through these rules in more detail.
The tracker’s court-order data shows the same turn. Counting each entry as issued, the tracker has 66 rules, orders and notices dated April 1 through September 30, including four dated only by month. Of those, 31 required filers to disclose AI use, and 29 of the 31 were still in force on September 30, 2026. Of the 68 dated 2023 and 2024, 44 did. Entries that include a step to verify AI-generated content, already common, became more so: 55 of the 66 recent entries, against 51 of the 68 earlier ones. The 55 include proposed rules, eight advisory notices or guidelines, and Iowa’s rule for judges. Four recent entries make signing or filing a document a certification that cited authority is genuine, whether or not AI was used. None of the earlier entries did. The four are the Florida and Indiana rules, the proposed Texas rule, and a Western District of North Carolina order, In re: Use of Artificial Intelligence, No. 3:24-mc-104, Doc. 2 (W.D.N.C. July 29, 2026) (tracker).
For a firm that files in several courts, the change is in what the obligation asks. A disclosure order asks which tools were used. The newer rules ask whether the citations are real, and several make the signature itself the representation. Whether a firm’s current review would support that representation is a question for the firm and its counsel.
Federal chambers kept writing their own orders
State rules do not reach federal courts, which account for 32 of the 66 recent entries. Judge Anne-Leigh Gaylord Moe (M.D. Fla.) is the clearest example (tracker). She entered a standing order on 2026-05-27 and, by September 30, 2026, had entered it in many of her other cases, one docket at a time.
Where it is entered, every filing must carry one of two certifications, verbatim and under penalty of perjury. One says AI was not used “in any way.” The other describes how it was used and states that the filer personally reviewed Rule 11 and personally read every case. Traditional Westlaw, LexisNexis and search-engine research is not AI use that must be disclosed. The order permits “no other form of certification,” and a filing without one may be stricken.
AOSC26-12 binds Florida state courts only, so the state rule left this federal order untouched. As of September 30, 2026, a firm with matters on both sides of the courthouse line in Florida worked under two different regimes. The court-order tracker lists the other chambers orders from the period.
Some chambers went the other way. Effective 2026-09-01, Senior Judge Jane J. Boyle (N.D. Tex.), by a requirement posted on her judge page (tracker), barred AI-generated text from attorney filings in her cases: “No text generated by artificial intelligence may be used in any portion of any pleading, motion, brief, or other filing made by an attorney before Judge Boyle.” As posted on September 30, 2026, the requirement called for a signed certificate on every filing that “the text was generated in its entirety by a human attorney,” and said that using AI to identify errors “must be disclosed under Local Rule 7.2(f).”
Federal rulemaking: a Rule 11 proposal, and a recommendation to drop it
On 2026-05-29 Magistrate Judge Patricia D. Barksdale of the Middle District of Florida suggested a fifth certification for Federal Rule of Civil Procedure 11 (Suggestion 26-CV-21). Her text: “the legal authorities exist and are accurately cited.” She asked for a subsection “akin to what the Florida Supreme Court has done.”
The agenda book for the Advisory Committee on Civil Rules’ meeting scheduled for October 21, 2026 (Tab 15), posted September 28, recommended dropping the suggestion. Its memo reasoned that “for now it seems that current Rule 11 is sufficient to support appropriate judicial action.” That was a recommendation to the committee, not its decision; the committee had not acted on it as of September 30, 2026. The suggestion appears on the timeline.
The Tenth Circuit was moving the other way within its own rules. Its proposed 10th Cir. R. 46.5(B)(4) (tracker) would have a filer certify that any document prepared with generative AI was reviewed by a human who independently verified it. As of September 30, 2026, the court was accepting public comments through October 18, 2026.
Sanctions: 188 decided matters in six months
As coded in the tracker on 2026-10-01, 188 matters were decided April 1 through September 30 in which a court or disciplinary tribunal issued a final decision on errors attributed to AI in filings or other submissions, such as fabricated citations. Pending show-cause matters are excluded. By month: 35 each in April and May, 29 in June, 36 in July, 32 in August, and 21 in September. In 63 of the 188, the court’s response was a warning.
Read those numbers as a floor. Most of this period’s entries were added in the September catch-up, and the latest months are the least complete, because decisions reach dockets and databases weeks after they issue.
Of the 188, 63 carry a dollar award. The median award was $1,750. The largest among them was $85,056 in fees and costs, which the Southern District of New York ordered a plaintiff and its counsel to pay (LiveVideo.AI Corp v. Redstone, No. 24 Civ. 6290 (DEH), ECF No. 211 (S.D.N.Y. June 9, 2026); tracker). The motions behind it violated a filing injunction, and the court found they appeared to have been drafted by AI and contained hallucinated quotations. Next was $45,763 in fees, set by the Western District of Tennessee after it imposed Rule 11 sanctions on a plaintiff law firm for misusing AI in court filings (Reaves Law Firm, PLLC v. Baker, Donelson, Bearman, Caldwell & Berkowitz, PC, No. 2:25-cv-2623, ECF No. 43 (W.D. Tenn. July 10, 2026); tracker).
Two orders from the period show what courts looked at.
Both sides of one case (N.D. Miss.). In a breach-of-contract suit against the City of Aberdeen, Senior Judge Sharion Aycock found on 2026-06-08 that briefs from the plaintiff’s counsel and from the city’s counsel cited authorities the court could not locate. The attorneys admitted the citations came from unverified AI use. The court held that all four had violated Rule 11. Two were barred from appearing in any case in the Northern District of Mississippi for two years and fined $2,500 and $3,500. The other two were disqualified from the case and fined $1,000 each. The order speaks to these four lawyers and this docket. (Withers v. City of Aberdeen, No. 1:24-CV-218-SA-RP, ECF No. 123 (N.D. Miss. June 8, 2026); tracker)
A six-month suspension (9th Cir.). On 2026-06-03, reviewing an immigration ruling, the Ninth Circuit sanctioned two attorneys at an Orange County, California firm $2,500 each and suspended both from practice before the court for six months. Their opening brief cited two cases that do not exist and attributed quotations to real opinions that do not contain them. Law-school graduates not yet licensed had written the legal arguments. One attorney attested that the two lawyers “do not normally vet citations used by the Brief writer during our review.” The court called that “an extraordinary confession.”
The court tied the suspension to the attorneys’ “repeated failure of candor.” Their motion to correct had called the fake cases “typographical errors.” At oral argument, one attorney said AI “was not used.” Had the attorneys disclosed in that motion that AI was used against firm policy, and apologized for failing to check the brief, the court wrote, “lesser sanctions may have been warranted.” Two facts in the order concern firm process rather than the tool: who checked the citations before signature, and what the firm told the court when asked. Firms reviewing their own drafting workflow may wish to consider how each would read in their filings. (Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026); tracker)
Judges, too: Iowa adopts rules for judicial AI use
By order signed 2026-07-21 by Chief Justice Susan Larson Christensen, the Iowa Supreme Court adopted Chapter 27 of the Iowa Court Rules (Iowa Ct. R. ch. 27) (order; tracker), “Judicial Use of Artificial Intelligence Tools,” effective 2027-01-01. It governs judicial officers and their adjudicative assistants. A judicial officer “must not delegate responsibility or defer to” AI work product. Every factual or legal source in that work product must be independently verified, and the state court administrator must publicly post the tools approved for use. The chapter imposes no obligations on attorneys or parties, and this post draws no inference from it about what Iowa courts will expect of filers.
A statute, not a court rule
California has written disclosure into statute. On 2026-09-30 the Governor approved SB 574 (Stats. 2026, ch. 858). New Business and Professions Code section 6068.1 covers attorneys who use generative AI to assist in the practice of law. They must take reasonable steps to verify its outputs, “including, but not limited to, the accuracy of all case and statutory citations.” They must also “[d]isclose the use of generative artificial intelligence to the court for all documents submitted to the court.” The bill also amends Code of Civil Procedure section 128.7: a filed paper “shall not contain any citations that an attorney responsible for submitting the pleading has not personally verified.” It directs the Judicial Council to revisit Standard 10.80 and adds a section on arbitrators. As a non-urgency statute it takes effect 2027-01-01.
That runs against the court-rule trend described above. Where courts have been trading disclosure for certification, California’s legislature will require both from January 1, 2027. An attorney who uses generative AI must disclose that use to the court for documents submitted to it, and every citation in a filed paper must be one the responsible attorney has personally verified. Firms with California matters may wish to consider how they would document AI use in court filings before that date. (California · SB 574 on the timeline)
State bars
California. The State Bar’s Board of Trustees approved a revised Practical Guidance for the Use of Generative Artificial Intelligence on 2026-05-14, replacing the 2023 version. A separate package would amend the Comments to Rules of Professional Conduct 1.1, 1.4, 1.6, 3.3, 5.1 and 5.3 to address generative AI. Under Rule 1.0(c), comments “are not a basis for imposing discipline.” The package went through public comment periods that closed 2026-05-04 and 2026-08-06, and it had not reached a Board vote as of September 30, 2026. SB 574’s duties, described above, are statutory and do not depend on it. (California)
Rhode Island. By order entered 2026-06-17, In re Amendments to Article V, Rule 1.1 of the Supreme Court Rules (Professional Conduct) and Adoption of Interim Generative AI Guidelines (R.I. June 17, 2026), the Supreme Court of Rhode Island amended Comment [6] to Rule 1.1: a lawyer should keep abreast of changes in the law and its practice, “including the benefits and risks associated with existing and developing technology.” The same order adopted interim guidelines on generative AI for lawyers and for judicial officers. The order says the guidelines “neither amend” the Rules of Professional Conduct “nor carry the force of law.” The lawyers’ guidelines say that before filing work product created with a generative AI tool, “a lawyer must review all such outputs for accuracy and correctness, verify citations and statements of fact or law.” (Rhode Island)
By the numbers
- 790 case entries on the tracker as coded on 2026-10-01. Of these, 606 are decided matters: pending show-cause matters and entries recorded as no sanction are excluded, and warnings are included.
- 244 court-order entries as coded on 2026-10-01, 222 of them in force or adopted.
- A third-party compilation, Damien Charlotin’s AI Hallucination Cases database, listed 2,097 decisions worldwide and 1,432 from the United States as of September 30, 2026. It includes matters involving self-represented litigants and some where AI use was alleged but not confirmed.
What the orders ask for
The rules and orders in this post converge on a short list: citations that exist, cited accurately, and a signature that vouches for them. How a firm documents the review behind that signature is its own choice. As a practical inference rather than a term of any rule, a firm that keeps a record of its citation checks has the readiest answer if a court asks.
At least one malpractice carrier has asked about AI on its application. AmTrust’s Lawyers Professional Liability Application (form LPLPRO-APP-01 0523) lists, among its internal procedures, “Allow the use of Artificial Intelligence software to draft documents. If checked, please attach description.” A firm’s own carrier is the place to ask what its next renewal will require. The state tracker shows what applies in each jurisdiction, and the carrier tracker follows what carriers have published.
Corrections and anything the tracker missed are welcome through the contact page.