Skip to content
Menu
Patent pending

The sanctions record · Missouri Court of Appeals, Eastern District, 2026

State ex rel. Schnell v. Richardson

State of Missouri ex rel. Steven Schnell v. Honorable Rebecca Richardson, No. ED114731, Writ Division Four · Presiding Judge John P. Torbitzky, with Judges Lisa P. Page and Michael S. Wright concurring · filed 8 September 2026

Page verified 9 September 2026. Opinion read in full, six pages; every quotation is from that opinion. Source: the opinion as filed, obtained as a PDF from the AI Hallucination Cases document store on the day the Missouri courts’ host and Justia refused automated retrieval; the Justia reprint is linked below.

AI use was not the violation. Filing unverified work was. The Missouri Court of Appeals ordered relator’s counsel to pay $1,750 after she admitted using generative AI to research legal authorities and failing to verify two citations that she “now admits do not exist”. The court separated the tool from the duty in one paragraph: “this Court is not sanctioning Conley because she used generative AI. The use of AI is only incidental to the sanction.” The sanction is for filing pleadings “that cited cases that do not exist for unsupportable propositions”, which the court said is “no different than if Conley signed her name to work prepared by someone else without verifying the accuracy of claims therein”.

The record

  • CourtMissouri Court of Appeals, Eastern District, Writ Division Four
  • PanelPresiding Judge Torbitzky, for the court; Judges Page and Wright concurring
  • CaseNo. ED114731, an original writ in mandamus seeking transfer of venue in a termination of parental rights and stepparent adoption matter
  • Filed8 September 2026
  • SanctionedRelator’s counsel, Jenna Rohr Conley
  • ToolGenerative AI, not named. Counsel’s response to the show cause order “stated that she used generative artificial intelligence to research legal authorities supporting her position but failed to verify the accuracy of the citations before submitting them to the Court”
  • False citationsTwo. The citations “did not correspond to the cases cited”; the court found cases bearing the same names, “but not at the citations provided by Conley”, and those cases “do not stand for the proposition identified”
  • Amount$1,750, paid to the petitioners in the underlying action as reasonable attorney’s fees for preparing suggestions in opposition to the writ petition
  • AuthorityMissouri Supreme Court Rule 55.03(b) and 55.03(c)
  • DisciplineNo referral ordered. The court “makes no decision regarding whether Conley’s conduct violated any rule”; that rests with the Supreme Court of Missouri (footnote 3)
  • MeritsWrit petition denied. The court found no authority for the proposition that all cases involving one family must be heard by a single court

What the court found

The two citations were offered for the proposition that every case involving one family must be decided by the same judge. On that representation of the law the court ordered the respondent to file suggestions in opposition. When it read the pleadings together, it found the problem: “The citations did not correspond to the cases cited. And the cases at the cited locations did not involve the subject matter at issue in the writ petition.”

The court described counsel’s response to its show cause order as “earnest and apologetic”, and sanctioned anyway. The reason is the court’s own reliance: “Worse still, this Court acted on Conley’s misrepresentations and ordered Respondent to file suggestions in opposition.” The $1,750 is measured by that wasted work.

Rule 55.03(b) makes every signature a certification, after “an inquiry reasonable under the circumstances”, that the legal contentions “are warranted by existing law”. The court found two violations. First: “for a claim to be ‘warranted by existing law,’ the law being used to support it must actually exist.” Second, the prefiling inquiry was not objectively reasonable, because the tendency of generative AI to fabricate case law and citations “is well documented”, citing Kruse v. Karlen (Mo. App. 2024) and Mata v. Avianca, and its tendency to misstate holdings and invent testimony from a record, citing State v. Coleman (Ohio Ct. App. 2026).

“Given the well-known tendency of generative AI models to make fabrications, it is objectively unreasonable for any attorney or self-represented party to fail to verify AI’s accuracy. For this reason alone, sanctions are appropriate.”

Then the paragraph that makes this a page rather than a row:

“To be clear, this Court is not sanctioning Conley because she used generative AI. The use of AI is only incidental to the sanction. There is nothing inherently problematic from a briefing perspective with the use of generative AI. Rather, the sanction is appropriate because Conley filed pleadings in this Court that cited cases that do not exist for unsupportable propositions. In this respect, it is no different than if Conley signed her name to work prepared by someone else without verifying the accuracy of claims therein.”

The court went on to say that generative AI in legal practice “is now commonplace” and “can be profoundly beneficial”, listing discovery review, summarising, outcome insight, research and drafting, before adding that careless use “carries great risks”. It named the professional conduct rules that careless use implicates, Rule 4-1.1 on competence including “the benefits and risks associated with relevant technology”, Rule 4-3.1, Rule 4-3.3, Rule 4-5.4 and Rule 4-8.4, while deciding none of them, and directed every Missouri attorney to Missouri Informal Opinion 2024-11 of the Office of Ethics Counsel, dated 25 April 2024.

Why it matters

The amount is small. The proposition is not. An appellate court has located the duty in the signature rather than in the tool, and has said that the AI is “incidental”. A lawyer who signs AI output without checking it is in the same position as a lawyer who signs an associate’s draft without reading it. The question the court asks is not which tool was used but what the signer did before signing.

Two other features travel. The standard is objective and, in the court’s view, settled: given what is now known about fabrication, failing to verify is unreasonable on its own, without any inquiry into intent. And the harm the court prices is not the false citation in the abstract but the court’s reliance on it: it issued an order on the strength of the authorities, and the other side had to answer.

The opinion also carries this record forward. It quotes State v. Coleman, the Ohio appellate decision on fabricated transcript quotations, which has its own page here, and it cites Kruse v. Karlen, Missouri’s earlier AI citation case.

The lesson for the next filing

The court did not ask whether AI was used. It asked whether anyone verified the citations before the signature went on the page, and the answer, on counsel’s own account, was no. Nothing in the record could show otherwise, because nothing had been recorded.

Regulayer creates checkable evidence of who approved exact words and when: which AI was used, what a named person reviewed, and the document it was sealed to, made at the time and verifiable free by the court without trusting the person who made it. It does not do the checking. It shows that the checking happened, and who stood behind it.

Generate the certification →  ·  Verify a sealed record free →

Sources

Related

Verified 9 September 2026 against the opinion, read in full. Information, not legal advice.

Information, not legal advice. Every entry is verified against the court’s own document or contemporaneous reporting; where a source is reporting rather than the document, we say so.