Part of the AI Regulation News hub.
An Ohio appellate court treated a pro se appellant's disclosure that he used generative AI to write his brief as establishing that he could not tell whether the arguments were viable
Disclosure is the remedy almost every court rule and bar opinion reaches for. This appellant disclosed, unprompted, and the court's response was that because he is not an attorney, his disclosure established he could not evaluate what the tool had produced.
Bottom line: A decision affirming a minor misdemeanour speeding conviction. In the course of it, the court recorded that the appellant had notified it that he used generative AI to create his brief, and held that since he is not an attorney he cannot determine whether AI has produced viable legal arguments.
Who this affects: Anyone drafting AI disclosure rules for courts, tribunals or regulators, and self-represented litigants relying on AI tools. The reasoning cuts against the assumption that disclosure cures the problem.
Issue date: Decided 3 August 2026, reported at 2026-Ohio-2993.
What changed: No rule changed. The contribution is the treatment of disclosure: the court read it as an admission bearing on the filer's capacity to verify, not as a mitigating step.
Analysis: Most disclosure regimes assume a filer who can vouch for what they file, so that disclosure adds transparency on top of an existing duty of accuracy. The court identifies the gap: a non-lawyer disclosing AI authorship is disclosing that the arguments came from a source they are not equipped to evaluate. Disclosure then documents the problem rather than curing it.
Primary sources: State v. Davis, 2026-Ohio-2993, opinion PDF · Same opinion on the Ohio Supreme Court reported opinions site
- Instrument
- State v. Davis, opinion and judgment entry
- Citation
- 2026-Ohio-2993; Case No. 25 CO 0050
- Court
- Court of Appeals of Ohio, Seventh Appellate District, Columbiana County, on appeal from the Columbiana County Municipal Court, Case No. 2025 TRD 002564
- Panel
- Waite P.J., Hanni and Dickey JJ. Opinion by Waite P.J.; Hanni and Dickey JJ. concurring.
- Jurisdiction
- Ohio, United States
- Status
- Decided 3 August 2026. Final judgment entry.
- Outcome
- Judgment affirmed. All six assignments of error overruled. Costs taxed against the appellant.
- Underlying matter
- Appeal from a bench trial conviction on a minor misdemeanour speeding citation
- Sourcing note
- The Ohio reported opinions host was unreachable from our network, a transport outcome rather than a statement about the document. The copy used is CourtListener's mirror, verified byte-identical by SHA-1 against the hash CourtListener publishes.
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://storage.courtlistener.com/pdf/2026/08/03/state_v._davis.pdf
What the appellant disclosed
The appeal concerned a bench trial conviction on a minor misdemeanour speeding citation, with six assignments of error. All were overruled and the judgment was affirmed.
Before reaching them the court dealt with the brief itself. It found the brief non-conforming and well outside the limits set by the Rules of Appellate Procedure: set in a small font, single spaced, and if converted to 12-point double-spaced would well exceed 35 pages and 9,000 words, with a word count over 12,000.
It then recorded the disclosure. The appellant had notified the court that he used generative artificial intelligence to create his brief. That notification appears to have been volunteered rather than extracted, which is what makes the court's response worth reading closely.
The response was not to thank him for the candour or to apply a sanction. It was to draw a conclusion about capacity: since the appellant is not an attorney, he cannot determine whether AI has produced viable legal arguments.
The burden the court identified
The court took the reasoning one step further, and this is the sharpest sentence in the passage. Since the appellant had not tested most of his arguments at the trial court level, the court said he seeks for this Court to determine the legitimacy of his AI documents and arguments.
That names the actual imposition. An untested AI-generated argument arrives at an appellate court with no filter applied by anyone: not by a lawyer, not by the filer, and not by a trial court that has already considered and ruled on it. Verification then falls to the appellate panel by default.
The court paired this with the settled rule that pro se litigants are presumed to have knowledge of the law and legal procedures and are held to the same standard as litigants represented by counsel, citing State ex rel. Neil v. French, and stated that the appellant's not understanding the rules or the pertinent law does not excuse him.
So the two propositions sit together without contradiction. The appellant is held to a lawyer's standard, and simultaneously is found unable to assess whether the tool that wrote his brief produced viable arguments. The gap between those is exactly the exposure a self-represented AI user carries.
The hallucination the court flagged in passing
The opinion also records a probable fabrication, and does so with more care than the topic usually attracts.
Addressing a cited case, the court said the issue in that case was whether the defendant was being held solely on the pending charge, and specifically whether an immigration detainer nullified a statutory triple count provision, which was not an issue in the present matter. It concluded that the case does not discuss the topic the appellant raises and does not mention indefinite tolling.
It then offered an explanation rather than a finding: the citation to this immaterial case may have been as a result of a hallucination by the appellant's AI queries.
Note the shape of that error. The case was real and correctly cited; it was simply about something else entirely. That is a third distinct failure mode, alongside wholly fabricated cases and real cases quoted with invented language, and it is the least likely to be caught by any check that stops at confirming the authority exists.
Why this matters to disclosure policy
Disclosure requirements are the most common regulatory response to AI in filings. Court practice directions, bar guidance and tribunal rules increasingly ask filers to state whether generative AI was used, on the theory that transparency lets the tribunal calibrate its scrutiny.
This decision exposes an assumption inside that theory. Disclosure regimes are generally built for filers who can vouch for their filings, so the disclosure supplements an existing duty of accuracy. Where the filer is a non-lawyer, the court here concluded that no such capacity exists.
The consequence is that disclosure converts into evidence of the very deficiency it was meant to manage. The appellant told the court he used AI; the court concluded he therefore cannot evaluate the result and is asking the court to do it for him.
That is not an argument against disclosure rules, and it should not be read as one. It is a reason to pair them with something else. A rule that asks whether AI was used, without also asking whether the filer verified the authorities and how, collects the fact while leaving the risk exactly where it was.
The line to take from this is the court's own: because the appellant is not an attorney, he cannot determine whether AI has produced viable legal arguments, and having not tested those arguments below, he was in substance asking the appellate court to determine the legitimacy of his AI documents. For anyone drafting AI rules, that is the weakness in disclosure as a standalone remedy. Disclosure assumes a filer who can vouch for the filing; where the filer cannot, the disclosure records the problem and shifts verification onto the tribunal. A rule that asks whether AI was used should also ask whether the authorities were verified and how. Two secondary points are worth keeping: the court held the appellant to the same standard as a represented litigant notwithstanding his lack of understanding, and it identified a third failure mode, a real and correctly cited case that simply does not address the issue, which no existence check will catch.
Source File
https://storage.courtlistener.com/pdf/2026/08/03/state_v._davis.pdf
Open the opinion PDF and confirm four things: the caption citing State v. Davis, 2026-Ohio-2993, Seventh Appellate District, Columbiana County, Case No. 25 CO 0050, dated 3 August 2026 before Waite, Hanni and Dickey JJ. with judgment affirmed; paragraph 14, recording the non-conforming brief and that the appellant notified the court he used generative artificial intelligence to create his brief and that as a non-attorney he cannot determine whether AI has produced viable legal arguments; paragraph 15, holding pro se litigants to the same standard as represented litigants; and paragraph 21, suggesting that an immaterial citation may have resulted from a hallucination by the appellant's AI queries.
Appellant has also notified this Court that he used generative Artificial Intelligence (AI) to create his brief. Since Appellant is not an attorney, he cannot determine whether AI has produced viable legal arguments. Court of Appeals of Ohio, Seventh Appellate District, State v. Davis, 2026-Ohio-2993, paragraph 14
FAQ
Did the appellant hide his AI use?
No. The court records that he notified it that he used generative AI to create his brief. The disclosure appears to have been volunteered.
What did the court conclude from the disclosure?
That because he is not an attorney, he cannot determine whether AI has produced viable legal arguments, and that having not tested most of his arguments in the trial court he was seeking to have the appellate court determine the legitimacy of his AI documents and arguments.
Was he sanctioned?
No sanction is recorded for the AI use. The appeal was resolved on its merits, all six assignments of error were overruled, the judgment was affirmed and costs were taxed against him.
Does being self-represented lower the standard?
No. The court restated that pro se litigants are presumed to have knowledge of the law and legal procedures and are held to the same standard as represented litigants, and that not understanding the rules or the law does not excuse a filer.
What was the hallucination the court mentioned?
A cited case that was real and correctly cited but addressed an unrelated issue, namely whether an immigration detainer nullified a statutory triple count provision. The court said the citation may have resulted from a hallucination by the appellant's AI queries.
Does this mean disclosure rules do not work?
It does not say that. It shows that disclosure alone assumes a filer capable of vouching for the filing. Where that capacity is absent, disclosure records the deficiency rather than curing it, which is an argument for pairing disclosure with a verification requirement.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.