Background
Matthew Brett Reeves, an Oklahoma-licensed attorney in good standing since 2003, was sanctioned by the U.S. District Court for the Northern District of Alabama (July 23, 2025) after he used ChatGPT to generate five case citations and inserted them — unverified — into two filed motions in Johnson v. Dunn. The plaintiff identified the problem: one citation existed but stood for an entirely different proposition, two cases did not exist at all, one citation resolved to an unrelated maritime case, and one was a 1939 Alabama traffic-offense decision. Reeves admitted in the federal proceeding that the citations were “completely made up” by ChatGPT and that he had not run a single check on any of them. The federal court characterized his conduct as “complete and utter disregard for his professional duty of candor” and “recklessness in the extreme,” noting that his firm had a written policy requiring practice-group-leader approval before using AI for legal research — a policy Reeves violated.
Reeves self-reported the federal sanction to every state where he was licensed. Reciprocal discipline followed: a public reprimand from Texas (November 4, 2025), a public censure from Tennessee (December 15, 2025) for violations of diligence, expediting litigation, and misconduct rules, and a public reprimand from the Alabama State Bar (March 6, 2026) for violations of competence, candor toward the tribunal, and misconduct rules. The Oklahoma Bar Association then transmitted the disciplinary records to the Oklahoma Supreme Court pursuant to Rule 7.7 of the Rules Governing Disciplinary Proceedings (RGDP), recommending a reciprocal public reprimand.
Reeves waived a hearing in Oklahoma. He contested the sufficiency of the Tennessee and Alabama proceedings but submitted no transcripts or independent evidence to rebut the prima facie case established by the out-of-state adjudications. In mitigation, he cited more than twenty years of discipline-free practice, prompt self-reporting, full cooperation, lack of selfish motive, and post-sanction work with opposing counsel to develop an AI-education program for law students.
The Court’s Holding
The Oklahoma Supreme Court publicly reprimanded Reeves, imposing reciprocal discipline consistent with the sanctions imposed in Texas, Tennessee, Alabama, and the federal court. Exercising its exclusive, original, and nondelegable jurisdiction over attorney discipline, the court held that the out-of-state adjudications constituted prima facie evidence of misconduct under RGDP 7.7(c), and that Reeves had not met his burden of showing the findings were unsupported or insufficient to ground discipline in Oklahoma. The court found that his conduct violated Oklahoma Rules of Professional Conduct (ORPC) 1.3 (diligence), 3.2 (expediting litigation), 3.3 (candor toward the tribunal), and 8.4(d) (conduct prejudicial to the administration of justice), as well as RGDP 1.3.
On the candor issue, the court rejected Reeves’s argument that he lacked the “knowing” mental state required by ORPC 3.3. Citing State ex rel. OBA v. Krug, 2004 OK 28, the court reasoned that knowledge may be inferred from circumstances. Reeves knew his firm prohibited unauthorized AI use precisely because AI fabricates plausible-looking citations; he knew he used ChatGPT without approval; and he knew he never verified a single result. A lawyer who knowingly uses AI to generate citations and knowingly forgoes verification has knowingly made a false statement to the tribunal.
The court also cited its own administrative order, In re: Generative Artificial Intelligence, SCAD-2026-17, for the proposition that “[a]ll persons filing documents in Oklahoma courts are responsible for the content and accuracy of the documents.” It declined to impose greater discipline than the other jurisdictions, crediting the mitigating factors Reeves presented, while emphasizing that the primary goals of attorney discipline — protecting the public, the judiciary, and the profession, and deterring future misconduct — were served by a public reprimand.
Key Takeaways
- An attorney who uses generative AI to produce citations and files them without any independent verification commits a knowing false statement to the tribunal under ORPC 3.3, even absent an intent to deceive, because knowledge may be inferred from the circumstances of deliberate non-verification.
- Out-of-state disciplinary adjudications are prima facie evidence of misconduct under Oklahoma RGDP 7.7(c); an attorney who does not contest those proceedings and submits no countervailing evidence cannot later challenge their factual or analytical sufficiency in Oklahoma.
- Violating a law firm’s own AI-usage policy — implemented specifically to guard against AI hallucinations — is a significant aggravating circumstance that undermines any good-faith or lack-of-knowledge defense.
- This is the first reported Oklahoma attorney discipline case involving AI-fabricated citations; the court and the concurrence explicitly invoked deterrence, reminding attorneys of their oath and their Rule 11-equivalent obligations to make reasonable inquiry before filing.
Why It Matters
The decision is one of a growing line of authorities across multiple jurisdictions treating unverified AI-generated citations as professional misconduct rather than mere inadvertent error. Oklahoma’s analysis — that deliberate non-verification is tantamount to knowledge of falsity — gives the “knowingly” element of candor-toward-the-tribunal rules real teeth in the AI context, even when the attorney claims no intent to mislead. Paired with the court’s citation to its own standing administrative order on generative AI, the opinion signals that Oklahoma courts will treat citation verification as a non-negotiable baseline, not an aspirational best practice.
For practitioners nationwide, Reeves reinforces that firm AI policies carry disciplinary weight: ignoring an internal safeguard designed to prevent exactly the harm that occurred is strong evidence of the recklessness courts need to impose sanctions. Attorneys who rely on AI tools for any portion of court filings should treat independent verification through traditional legal research as mandatory — not optional — before signing any document submitted to a tribunal.