State v. Coleman (Ohio 2026): Attorney Sanctioned for AI-Hallucinated Quotes in Criminal Appeal | Advocate Prakhar

Case at a Glance

Background

Malikhi Jermaine Coleman was convicted of murder, discharge of a firearm near prohibited premises, and improperly handling a firearm in a motor vehicle, with firearm specifications. The convictions were affirmed by the 11th Appellate District in February 2025. Attorney William Norman, representing Coleman, filed an application to reopen the appeal in May 2025, submitting an affidavit sworn under penalty of perjury attesting to the truth of its contents.

The prosecution’s response noted that two direct quotes from the prosecutor — described by Norman as “legally inflammatory” — could not be located anywhere in the trial transcripts. Three separate prosecution staff members independently searched the record and failed to find the quoted language. The citation Norman provided — page 559 of the transcript — turned out to be the court reporter’s signature page, containing no statements of any kind.

After the prosecution moved for sanctions, Norman entered into a settlement agreement acknowledging that the quotes were fabricated. The court ordered full briefing on the gravity of the violations and determined a post-briefing hearing was unnecessary before issuing its sanctions decision.

The AI Issue

Attorney Norman’s application to reopen contained two prosecutor quotes presented as direct citations from the trial record. Neither quote existed. The State’s brief detailed specifically how the citations pointed to non-existent content. The court’s investigation revealed a pattern consistent with AI-generated legal drafting where the tool invented plausible-sounding quotes and attributed them to specific record pages. By swearing under oath to the accuracy of the brief’s contents, Norman compounded the violation. The core legal question was whether an attorney who files AI-hallucinated record citations under oath violates Rule 3.3 of the Ohio Rules of Professional Conduct (candor toward the tribunal) even when he claims he did not intentionally deceive the court.

What the Court Decided

• Filing fabricated record citations in a sworn application constitutes a violation of Ohio RPC 3.3, regardless of whether deception was intentional — the oath creates strict accountability for factual accuracy.
• The prosecution and court were forced to waste significant time searching for non-existent quotes, satisfying the prejudice element for sanctions under the court’s inherent authority.
• An attorney’s failure to verify AI-generated content before filing it under oath is itself the sanctionable act, independent of subjective intent.
• The settlement agreement admitting to fabricated citations was a mitigating factor but did not eliminate the need for public deterrence through formal sanctions.
• The court imposed a public reprimand and financial sanction, and referred the matter to disciplinary authorities, signaling that criminal defense practitioners bear the same verification duty as civil litigators.

“Appellee has searched through the trial transcripts as a whole for these statements, and has not located them… Appellant has therefore misled this Court. This is a sanctionable offense, and not one to be taken lightly.”

— State’s Opposition to Application to Reopen, adopted by the Court, 11th Appellate District

The India Angle

Indian Law Equivalent

Under Indian law, a lawyer who presents a false document or fabricated citation to a court may face contempt proceedings under the Contempt of Courts Act, 1971. Section 2(c) defines criminal contempt to include scandalizing the court or doing acts prejudicial to the administration of justice. The Supreme Court in In re: Vinay Chandra Mishra (1995) 2 SCC 584 held that an advocate’s conduct that obstructs or tends to obstruct the due course of judicial proceedings is punishable. AI-generated fabricated citations causing courts to waste judicial resources would qualify as interference with judicial administration.

Bar Council Rules

Bar Council of India Rules, Chapter II, Part VI, Rule 22 prohibits an advocate from making statements in pleadings that are not true to the best of the advocate’s knowledge and belief. Rule 14 requires advocates to uphold the dignity of the court and maintain the honour and dignity of their profession. Filing fabricated AI citations — especially after swearing an affidavit — would trigger disciplinary action under these rules before State Bar Councils and potentially the Bar Council of India under Section 35 of the Advocates Act, 1961.

Practical Advice for Indian Advocates

• Never file any brief, application, or affidavit containing case references, quotations, or record citations generated by AI without first independently verifying each citation against the original source document.
• When AI is used to draft appellate materials, maintain a verification log showing that each citation was cross-checked — this creates a defensible record if a court questions accuracy.
• If you discover a fabricated citation after filing, immediately bring it to the court’s attention under your duty of candour; self-correction substantially mitigates disciplinary exposure under BCI Rules.

Quick Takeaways

• Swearing to AI-generated fabrications in a criminal appeal is a severe Rule 3.3 violation.
• Courts waste time on non-existent citations — that alone establishes prejudice for sanctions.
• Verification before filing, not intent, is the standard courts will apply.

Deep Dive: AI Hallucinations in Criminal Appellate Practice

The stakes in criminal appellate work are uniquely high. When an attorney files an application to reopen a murder conviction, the client’s liberty depends on the accuracy of every factual claim. State v. Coleman illustrates how AI hallucinations create a category of professional failure that has no analogue in traditional legal drafting errors: the fabricated content is internally coherent, plausibly styled, and attribute-complete — making it nearly impossible to detect without actually locating the source.

What makes this case particularly instructive is the oath dimension. Norman did not merely file a brief with wrong citations; he swore an affidavit attesting to the truth of allegations built on those citations. This transformed an already serious error into a potential perjury situation, and the court’s analysis focused on the incompatibility between sworn statements and unverified AI output. The lesson is structural: if an advocate is going to swear to the accuracy of a document’s contents, that document cannot contain any element whose accuracy the advocate has not personally confirmed.

The Ohio court’s approach also resolves a question that has divided other tribunals: does intent matter? The settlement agreement implicitly acknowledged that Norman did not intend to deceive, but the court found that this did not eliminate sanctions. The duty under RPC 3.3 is forward-looking and prophylactic — it is satisfied only by a reasonable pre-filing inquiry, not by post-discovery regret. This aligns the AI hallucination sanction standard with the broader Rule 11 framework applied in federal courts: objective reasonableness of the filing inquiry, not subjective bad faith.

For Indian practitioners arguing before high courts and the Supreme Court where verification of record references is equally critical, the Coleman case is a direct precedent analog. India’s appellate system relies heavily on counsel’s representation of the record’s contents, and a lawyer who presents fabricated transcript quotes in a criminal revision petition or special leave petition creates precisely the same institutional harm — wasted judicial time, potentially wrongful denial of a client’s legal remedy, and erosion of court confidence in counsel. The path to safety is not avoiding AI, but implementing a disciplined verification workflow before any AI-assisted filing reaches a tribunal.