State of Oklahoma ex rel. Oklahoma Bar Association v. Reeves

What sanction did the court impose?

In an opinion by Justice Darby, with all nine justices concurring, the Supreme Court of Oklahoma publicly reprimanded Reeves. It held that the out-of-state adjudications are prima facie evidence under RGDP 7.7(c) and that the burden was his to show those findings unsupported or insufficient; he waived a hearing and "chose to not file any evidence" to meet it. It rejected his argument that he did not knowingly misstate the law, holding that knowledge may be inferred from circumstances: he alone used AI to find the citations, knew he had used it, and knew he had not verified what he added. The Court concluded that presenting false citations obtained through generative AI violated ORPC 1.3, 3.2, 3.3, and 8.4(d) and RGDP 1.3. Reeves argued in mitigation that he had no prior discipline in over twenty years, self-reported, acted without dishonest motive, and now helps educate law students about AI risks. Justice Gurich, joined by Justice Combs, concurred separately to stress the duty to verify each citation and the attorney's oath. As issued, the opinion carried the notice "THIS OPINION HAS NOT BEEN RELEASED FOR PUBLICATION. UNTIL RELEASED, IT IS SUBJECT TO REVISION OR WITHDRAWAL."

Why does State of Oklahoma ex rel. Oklahoma Bar Association v. Reeves matter for law firms using AI?

Reeves is the reciprocal-discipline sequel to Johnson v. Dunn, the Northern District of Alabama order that sanctioned Reeves for five ChatGPT-generated citations in two discovery motions. Oklahoma was the fourth licensing jurisdiction to act, after Texas, Tennessee, and Alabama, and the opinion is useful mainly for how the Court handled Reeves’s defense that he never knowingly misstated the law.

The Court rejected that defense by inference. Citing its decision in Krug and the definition of “knowingly” in ORPC 1.0(f), it held that knowledge “may be inferred from circumstances,” and it found the circumstances here sufficient: Reeves “was the only attorney in the case who used AI to find citations,” was the only one who knew he had used it, and knew that he had not verified what he added. From there the Court stated a general rule: “If a lawyer knowingly uses AI to generate citations or analysis for a pleading and knowingly does not verify the accuracy before including it in the pleading, then the lawyer has knowingly made a false statement.” The firm’s own policies, which required written approval for ChatGPT research and verification of AI output, supplied part of the knowledge the Court inferred.

The Court was careful not to condemn the technology. “There is nothing inherently problematic with the use of generative artificial intelligence in preparing legal materials,” it wrote, “so long as attorneys abide by their duty to protect client confidentiality and recall their sacred duty to verify the pleadings they sign.” What it condemned was the signature: “Signing pleadings that contain citations from generative AI that have not been verified shows a reckless disregard for the truth and an indifference to accuracy.” The discipline imposed was a public reprimand, as in Texas and Alabama; Tennessee had issued a public censure. In a concurrence joined by Justice Combs, Justice Gurich expressed the hope that the reprimand would deter others “from the careless and unprofessional use of AI.”