Unverified AI-Generated Citations as “Knowing” False Statements Under ORPC 3.3 (Reciprocal Discipline Under RGDP 7.7)

I. Introduction

STATE OF OKLAHOMA ex rel. OBA v. REEVES, 2026 OK 37, is an original attorney-discipline proceeding in which the Oklahoma Bar Association (OBA) sought reciprocal discipline against attorney Matthew Brett Reeves based on discipline imposed in multiple jurisdictions following a federal sanctions order. The underlying misconduct occurred in the United States District Court for the Northern District of Alabama and involved the filing of two motions containing fabricated case citations generated by a generative artificial intelligence tool (ChatGPT) that were not independently verified.

The key issues before the Oklahoma Supreme Court were (1) how Oklahoma’s reciprocal-discipline framework under RGDP 7.7 operates when other jurisdictions’ disciplinary orders contain varying degrees of factual detail; and (2) whether submitting pleadings containing unverified AI-hallucinated citations can constitute a violation of Oklahoma ethical duties—particularly candor toward the tribunal—even where the lawyer claims lack of subjective awareness that the citations were false.

The Court publicly reprimanded Reeves, marking the Court’s first reported discipline decision squarely addressing AI-fabricated citations and the professional duty to verify before filing.

II. Summary of the Opinion

The Court held that there is “nothing inherently problematic” with using generative AI to prepare legal materials, but emphasized that human diligence and review are required to ensure accuracy. The Court determined that filing pleadings containing AI-generated citations without verifying them reflects reckless disregard for truth and is inconsistent with an attorney’s duties.

Applying RGDP 7.7, the Court treated the federal sanctions order and the out-of-state disciplinary actions (Texas public reprimand, Tennessee public censure, and Alabama public reprimand) as the “charges” and as prima facie evidence of the acts described. Reeves did not supply transcripts or comparable evidence from those other disciplinary proceedings and waived a hearing in Oklahoma. The Court rejected his attempt to undermine the Tennessee and Alabama actions by pointing to their brevity, noting the federal sanctions order supplied the factual foundation and that a lawyer cannot benefit from choosing not to contest charges elsewhere.

On the merits, the Court reasoned that knowledge under ORPC 3.3 may be inferred from circumstances, and announced a particularly consequential rule for AI use: if a lawyer knowingly uses AI to generate citations or analysis for a pleading and knowingly does not verify accuracy before filing, then the lawyer has knowingly made a false statement.

The Court concluded the conduct constituted violations of ORPC 1.3, 3.2, 3.3, 8.4(d) and RGDP 1.3, and imposed a public reprimand.

III. Analysis

A. Precedents Cited

  • State ex rel. Okla. Bar Ass'n v. Hart, 2014 OK 96, ¶ 6, 339 P.3d 895, 898: cited for the Court’s “exclusive, original, and nondelegable jurisdiction” over attorney discipline. This frames the case as one in which the Court independently determines both misconduct and sanction, even when proceedings are reciprocal.
  • State ex rel. Okla. Bar Ass'n v. Cox, 2011 OK 73, ¶ 10, 257 P.3d 1005, 1008 (quoting State ex rel. Okla. Bar Ass'n v. Taylor, 2003 OK 56, ¶ 2, 71 P.3d 18, 21): used to reaffirm that the Oklahoma Supreme Court retains ultimate responsibility for whether misconduct occurred and what discipline is warranted. This matters in reciprocal cases: foreign discipline informs but does not dictate Oklahoma’s result.
  • State ex rel. Okla. Bar Ass'n v. Henderson, 1999 OK 29, ¶ 4, 977 P.2d 1096, 1098: central to the reciprocal discipline framework. The Court relied on Henderson to apply RGDP 7.7(c)’s rule that certified foreign discipline documents constitute the charge and are prima facie evidence of the acts described—shifting the burden to the respondent to show lack of evidentiary support or insufficiency for Oklahoma discipline.
  • State ex rel. Okla. Bar Ass'n v. Bransgrove, 1998 OK 93, ¶ 6, 976 P.2d 540, 541: used to reject Reeves’s effort to discount out-of-state discipline after he did not contest it there. Bransgrove supplies the principle that a respondent “may not deny the effect” of choosing not to contest charges in another jurisdiction by later complaining that no evidentiary hearing occurred.
  • State ex rel. Okla. Bar Ass'n v. Krug, 2004 OK 28, ¶ 12, 92 P.3d 67, 71: crucial to the Court’s treatment of “knowledge” under ORPC 3.3. Krug clarifies that knowledge “may be inferred from circumstances,” enabling the Court to treat deliberate non-verification in the face of known AI unreliability and internal controls as satisfying the “knowingly” component of candor duties.
  • State ex rel. Okla. Bar Ass'n v. Patterson, 2001 OK 51, ¶ 5, 28 P.3d 551, 554: cited for the proposition that the Court may impose reciprocal discipline when another jurisdiction has already imposed discipline.
  • State ex rel. Okla. Bar Ass'n v. Kleinsmith, 2013 OK 16, ¶ 4, 297 P.3d 1248, 1250: cited twice for two related propositions: (1) the Court’s discretion in reciprocal discipline to impose the same, greater, or lesser sanction; and (2) the core aims of discipline (protect public/judiciary/profession and deter similar misconduct).
  • State ex rel. Okla. Bar Ass'n v. Benefield, 2005 OK 75, ¶ 21, 125 P.3d 1191, 1195: cited as part of the Court’s standard methodology—guidance by discipline imposed on other lawyers for similar misconduct—to promote proportionality.
  • In re: Generative Artificial Intelligence, SCAD-2026-17: not a bar-discipline precedent but a pivotal administrative/judicial policy reference. The Court invoked it for the categorical proposition that “All persons filing documents in Oklahoma courts are responsible for the content and accuracy of the documents,” reinforcing the decision’s verification mandate.

B. Legal Reasoning

1. RGDP 7.7 as a Burden-Shifting Framework in Reciprocal Discipline

The Court treated the sanctions order and the Texas/Tennessee/Alabama discipline documents as the “charge” and as prima facie evidence under RGDP 7.7(c). That prima facie effect is decisive: it shifts the burden to the respondent to provide evidence (including, expressly, transcripts from other jurisdictions) showing the findings were unsupported or insufficient for discipline in Oklahoma.

Reeves attempted to attack the Tennessee censure and Alabama reprimand as too conclusory to merit deference. The Court reframed this as a misunderstanding of RGDP 7.7: it is the respondent’s job, once prima facie evidence is established, to rebut it with proof. The Court emphasized that Reeves did not provide transcripts or other evidentiary material from the out-of-state proceedings and also waived a hearing in Oklahoma—choices that left the prima facie showing intact.

2. The Court’s AI-Specific Candor Rule: Deliberate Non-Verification as “Knowing” Falsehood

Reeves’s central mitigation argument against an ORPC 3.3 violation was that he did not “knowingly” make false statements because he lacked actual awareness that the citations were fabricated. The Court’s analysis did not treat this as a mere negligence question. Instead, it applied the inferential-knowledge approach recognized in State ex rel. Okla. Bar Ass'n v. Krug and in ORPC Rule 1.0(f), and relied heavily on contextual facts demonstrating that Reeves:

  • knew he used generative AI to produce citations;
  • knew he did not verify any of the citations;
  • knew his firm had explicit, repeated warnings and policies about AI unreliability and mandatory verification; and
  • filed the motions anyway, attaching his name and professional certification to the content.

The Court’s doctrinal move is explicit and forward-looking: “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.” This effectively operationalizes “knowledge” for AI-assisted lawyering by tying it to a conscious decision to bypass verification in a known risk environment.

3. Systemic Harm and the Integrity Rationale

The Court’s reasoning is grounded in institutional integrity rather than case-outcome prejudice. It characterized unverified AI-citation use as: (1) a waste of judicial resources and adversary costs, and (2) conduct that “damages the integrity and credibility of the legal system.” This is consistent with its ORPC 8.4(d) conclusion (conduct prejudicial to the administration of justice) and its emphasis that lawyers’ signatures represent verified legal support.

4. Discipline Selection: Deterrence and Proportionality

For sanctioning, the Court adopted the federal court’s deterrence framing that sanctions should make AI misuse “unprofitable,” correspond to the “extreme dereliction” reflected by sham citations, and communicate that made-up authorities have no place in court. Still, the Court selected public reprimand, aligning Oklahoma’s sanction with the respondent’s reciprocal outcomes and considering mitigating factors (no prior discipline, self-reporting, cooperation, remorse, remedial education efforts).

C. Impact

  1. Establishes Oklahoma’s first reported AI-citation discipline rule. The opinion supplies a clear, court-announced standard: AI is permitted, but verification is mandatory. The decision is likely to serve as Oklahoma’s baseline authority for future bar complaints involving AI-assisted research, drafting, or citation.
  2. Recalibrates “knowingly” for modern practice risks. By treating deliberate non-verification in the face of known AI unreliability as “knowing” falsity (via inferred knowledge), the Court reduces the viability of defenses framed as “I didn’t realize it was false,” where the attorney consciously bypassed checks.
  3. Strengthens reciprocal discipline as a practical enforcement mechanism. The decision reinforces that under RGDP 7.7, foreign discipline documents carry prima facie effect, and respondents must bring rebuttal evidence (not merely argument about brevity). This will matter as AI-related sanctions and bar actions proliferate across jurisdictions.
  4. Signals potential overlap with court-rule sanctions regimes. Justice Gurich’s concurrence connects professional discipline with attorneys’ sworn oath (5 O.S. § 2) and with 12 O.S. § 2011’s certification/reasonable inquiry obligations, suggesting that AI-citation failures can trigger both bar discipline and litigation sanctions.
  5. Professional compliance consequences for firms and individual lawyers. The Court’s reliance on internal firm AI policies as evidence of inferred knowledge implies that firm trainings, written policies, and warnings may become relevant evidence in future discipline cases—both as aggravation (ignoring controls) and potentially as mitigation (robust compliance and prompt remediation).

IV. Complex Concepts Simplified

  • Generative AI “hallucinations”: instances where an AI system produces confident-sounding but false information—here, wholly fabricated legal citations and propositions.
  • Duty of candor (ORPC 3.3): the obligation not to make false statements of fact or law to a court and to correct material falsehoods. This opinion stresses that candor includes not “turning a blind eye” to whether your own citations are real.
  • “Knowingly” and inferred knowledge: “Knowingly” can mean actual knowledge, but courts may infer it from circumstances. In practical terms, when a lawyer knows AI may fabricate authorities and chooses not to verify, the lawyer cannot easily avoid responsibility by claiming ignorance of the specific falsity.
  • Reciprocal discipline (RGDP 7.7): when Oklahoma considers discipline based on discipline imposed elsewhere. The other jurisdiction’s certified documents become the charge and prima facie evidence, and the lawyer must rebut them with proof if disputing them.
  • Prima facie evidence: evidence that is sufficient to establish a fact unless rebutted. Here, the out-of-state orders were enough to establish the misconduct unless Reeves produced evidence undermining them.
  • ORPC 8.4(d) (prejudicial to the administration of justice): misconduct that undermines the functioning of courts. The Court treated wasted resources, compromised credibility, and forced corrective litigation as core “administration of justice” harms.

V. Conclusion

STATE OF OKLAHOMA ex rel. OBA v. REEVES sets a durable, practice-shaping principle for Oklahoma lawyers: using generative AI is not prohibited, but filing AI-generated citations without independent verification is professional misconduct. By linking deliberate non-verification to “knowing” falsity under ORPC 3.3 (through inferred knowledge), the Court closes a foreseeable accountability gap in AI-assisted advocacy.

The decision also reinforces the strength of RGDP 7.7 in reciprocal matters: foreign discipline is presumptively valid evidence of the acts described, and a respondent must rebut with evidence—not simply critique the detail level of another jurisdiction’s order.

The case’s broader significance lies in its institutional message: courts will treat fabricated authorities—whether produced by “artificial or human intelligence”—as a direct threat to the integrity of adjudication, and Oklahoma attorneys remain fully responsible for the accuracy of what they sign and file.