⚡ Case Digest
STATE v. COLEMAN / IN RE WILLIAM B. NORMAN — Court of Appeals of Ohio, 11th Appellate District, March 20, 2026
A paralegal at Attorney William Norman’s firm used ChatGPT to draft an application to reopen a murder conviction — generating hallucinated prosecutor quotes that never existed. Norman filed the document under oath without reading it.
Why it matters: Courts will independently sanction AI misconduct even after parties settle privately — and the sanction package goes far beyond a fine.
Category: AI Hallucination & Sanctions | Jurisdiction: USA (Ohio) | Read time: 6 min
Case at a Glance
| Full Citation | State v. Coleman, 2026-Ohio-965 (11th App. Dist.); In re William B. Norman, Case No. 2024-A-0040 |
| Court | Court of Appeals of Ohio, Eleventh Appellate District, Ashtabula County |
| Date | March 20, 2026 |
| Category | AI Hallucination & Attorney Sanctions |
| Jurisdiction | United States — Ohio |
| AI Tool Used | ChatGPT (by non-attorney staff member) |
| Outcome/Sanction | $2,000 monetary sanction; referral to Ohio Office of Disciplinary Counsel; 2-year court-notification requirement; apology letters to affected prosecutors |
Background
Malikhi Jermaine Coleman was convicted of murder in Ashtabula County, Ohio. After the Eleventh District affirmed his convictions in February 2025, his attorney William B. Norman filed an application to reopen the appeal on May 19, 2025. The application — which Norman swore to under oath — alleged prosecutorial misconduct based on inflammatory quotes supposedly found in the trial transcript.
The State’s prosecutors could not locate either quote. After exhaustive searching by three staff members, they confirmed the quotes did not exist. The page Norman cited for one “legally inflammatory” statement was the court reporter’s signature page — blank of any prosecutor statement. The State moved for $6,000 in sanctions.
The AI Issue
The central question was whether Attorney Norman committed sanctionable bad faith by filing a sworn application containing ChatGPT-generated hallucinations — quotes that never appeared in the trial record — without personally reading or verifying the document his paralegal prepared. A secondary issue arose when the court discovered that Norman’s new AI policy document itself appeared to have been generated by ChatGPT, complete with unfilled placeholder fields like “[Insert Date].”
The court further learned that two months after the sanctions hearing, Norman filed another motion in a different case (State v. Saker) that again contained questionable citations — and one filing literally ended with a ChatGPT prompt: “Would you like me to draft the next argument section?”
What the Court Decided
- Norman acted in bad faith — not merely negligently — by filing a sworn document he had not read [bad faith under inherent sanctioning authority].
- A private settlement between parties does not strip the court of its independent power to sanction [inherent authority survives settlement].
- The $2,000 monetary fine (credit given for $2,000 already paid to the prosecutor’s office) established that courts will exercise inherent authority over AI misconduct as a matter of precedent.
- Referral to the Ohio Office of Disciplinary Counsel was mandatory because the conduct implicated Rules 1.1 (competence), 3.3 (candor), 5.3 (supervision of non-lawyers), and 8.4(c) and (d) [professional conduct rules].
- Norman must notify every court in which he appears for the next two years by serving a copy of this judgment on each presiding judge within 14 days of filing [transparency sanction].
- Norman must send written apology letters to the prosecutors whose statements he falsely attributed fabricated quotes to [reputational repair sanction].
“He didn’t catch the hallucinated quotes. That was wrong. That’s his responsibility.”
— Respondent’s own counsel, August 28, 2025 hearing, quoted in State v. Coleman, 2026-Ohio-965
The India Angle
Indian Law Equivalent
Section 209 of the Indian Penal Code (now Section 228 of the Bharatiya Nyaya Sanhita, 2023) criminalises making false claims in court. More directly, the Contempt of Courts Act, 1971 covers scandalising the court and making false statements — a lawyer who submits AI-fabricated citations could face criminal contempt proceedings. Section 18 of the Advocates Act, 1961 empowers the Bar Council of India to take disciplinary action for professional misconduct.
Bar Council Rules
Bar Council of India Rules, Chapter II, Part VI: Rule 14 requires an advocate to uphold the dignity and decorum of the court. Rule 22 prohibits an advocate from making statements the advocate knows to be false. Rule 15 demands that every advocate maintain fair and honest conduct. All three are directly engaged when AI hallucinations are filed as true legal authority.
Practical Advice for Indian Advocates
- Never file any document prepared by a paralegal, junior, or AI tool without personally reading and verifying every citation — an advocate’s signature is personal certification of accuracy.
- If AI assists in drafting, run every case citation through Manupatra, SCC Online, or Indian Kanoon before filing — hallucinated citations from foreign AI tools citing Indian cases are a real risk.
- Maintain a written AI usage policy for your office; if you do not have one, draft one now — courts are increasingly examining whether firms have governance structures around AI.
Quick Takeaways
- Signing without reading = bad faith, even when a paralegal made the error.
- A private settlement does not end the court’s power to sanction AI misconduct.
- Multi-pronged sanctions — fine + referral + notification + apology — are the new standard.
Deep Dive: Why the Ohio Court’s Sanction Package Sets a New Benchmark
State v. Coleman is not simply another AI hallucination case. It is a 65-page judicial treatise on why courts must respond to AI misconduct with multi-layered consequences rather than a single monetary slap. Judge Eugene A. Lucci’s opinion methodically dismantles every excuse Attorney Norman offered — and builds a framework that other courts across the country are likely to follow.
The facts are disturbing in their specificity. Norman’s paralegal used ChatGPT to prepare an application to reopen a murder conviction. The application claimed prosecutors made two inflammatory statements — a “duty to retreat” comment and a reference to the defendant’s AK-47 — which Norman cited as evidence of prosecutorial misconduct. The State’s response was unequivocal: three staff members spent hours combing the trial transcript and found neither quote. Page 559, which Norman specifically cited, was the court reporter’s signature page. There was simply nothing there.
When confronted at the August 2025 sanctions hearing, Norman’s own counsel delivered the most damning line: “He didn’t catch the hallucinated quotes. That was wrong. That’s his responsibility.” The court then discovered that Norman’s promised remediation — a new AI policy — was itself apparently generated by ChatGPT, containing generic boilerplate and unfilled placeholder fields. Two months after the sanctions hearing, a separate filing in State v. Saker contained a literal ChatGPT prompt embedded in the document: “Would you like me to draft the next argument section?” This pattern transformed Norman’s conduct from a one-time mistake into something the court found had the hallmarks of systemic failure.
The court’s sanction package is worth studying in detail. First, the $2,000 monetary fine: the court acknowledged that $2,000 was compensatory rather than punitive, referencing the 32.1 staff-hours the prosecutor’s office devoted to tracking down non-existent quotes. The court noted similar sanctions nationally — $5,000 in Mata v. Avianca, $3,000 and $1,000 in Wadsworth v. Walmart, $15,000 in Mid Central Operating Engineers — and deliberately calibrated its order within that range. Second, the disciplinary referral: this is mandatory under Ohio’s professional conduct rules whenever a court identifies misconduct raising fitness questions. Third — and most innovative — the two-year court-notification requirement: Norman must personally inform every presiding judge in every court where he files within 14 days. This transparent accountability mechanism ensures that future courts are not operating without knowledge of his prior AI failures. Fourth, the apology letters: directed at the prosecutors whose reputations were damaged by fictitious quotes attributed to them. The court framed this as addressing reputational harm beyond the judicial process itself. For Indian advocates, the lesson from State v. Coleman is that the stakes of AI-assisted hallucinations extend far beyond a monetary fine — they reach into professional licensing, public reputation, and the trust that courts place in officers of the court.