United States v. John C. Farris: Sixth Circuit Sanctions for Unverified AI-Generated Appellate Briefing and Reaffirms Nondelegable Duty of Candor and Citation Verification

Court: Court of Appeals for the Sixth Circuit
Date: April 3, 2026
Posture: Sentencing appeal stayed in practical effect while the court addresses counsel misconduct; court removes counsel and orders re-briefing with replacement CJA counsel.

I. Introduction

United States v. John C. Farris is a published Sixth Circuit opinion that does not decide the merits of the defendant’s sentencing challenge. Instead, it addresses appellate counsel’s misconduct in submitting briefs containing inaccurate quotations and misleading characterizations of precedent after using a generative artificial-intelligence tool to draft the briefs without adequate verification.

Parties. The United States (Plaintiff-Appellee) and John C. Farris (Defendant-Appellant). The focal actor for the opinion’s holdings is court-appointed counsel, Steven N. Howe, appointed under the Criminal Justice Act (“CJA”), 18 U.S.C. § 3006A.

Background issue on appeal (not resolved here). Farris pleaded guilty and received, at sentencing, a two-level leadership enhancement under U.S.S.G. § 3B1.1(c). He appealed that enhancement.

Key issues actually decided. The court addresses (1) how counsel used AI-assisted drafting, (2) how that use led to false quotations and misstatements of holdings, and (3) what remedies and sanctions are appropriate—particularly in a publicly funded, court-appointed context.

II. Summary of the Opinion

The Sixth Circuit determined that counsel filed two appellate briefs (principal and reply) containing multiple misrepresentations: quotations attributed to genuine authorities that did not exist in those sources, and descriptions of holdings in United States v. Washington, 715 F.3d 975 (6th Cir. 2013) and United States v. Anthony, 280 F.3d 694 (6th Cir. 2002) that were materially inaccurate.

After issuing a show-cause order, the court received counsel’s admission that he used Westlaw’s CoCounsel to generate first drafts and failed to properly verify the authorities and quotations, resulting in erroneous content. The court framed the problem as a breach of baseline ethical obligations—competence, candor, and supervision—regardless of whether AI is used.

Consequences ordered. The court:

  • Denied CJA compensation to counsel for time spent on the appeal.
  • Referred the matter to the Chief Judge for potential disciplinary proceedings under Sixth Circuit Local Rule 46.
  • Served the opinion on leadership and disciplinary authorities in the Eastern District of Kentucky and the Kentucky Bar Association.
  • Removed counsel from the case (by separate order), locked the filed briefs, appointed replacement CJA counsel, and reset briefing.

III. Analysis

A. Precedents Cited (and How They Shape the Decision)

1. Whiting v. City of Athens, --- F.4th ----, 2026 WL 710568 (6th Cir. 2026)

The court situates this case as part of an emerging line of Sixth Circuit encounters with AI-related litigation failures. By citing Whiting v. City of Athens, the panel signals that AI misuse is not hypothetical and that the Sixth Circuit is developing institutional responses. The citation functions less as a doctrinal anchor and more as a “notice” that the court is monitoring AI-driven breakdowns and expects lawyers to adapt immediately.

2. McCoy v. Ct. of Appeals of Wis., Dist. 1, 486 U.S. 429 (1988)

McCoy is invoked for the foundational proposition that attorneys must not misrepresent the law to the court—an expression of the duty of candor and competent advocacy. By relying on McCoy, the Sixth Circuit places AI-assisted errors within an older, technology-neutral framework: the problem is not novelty, but breach of professional obligation. In effect, the court treats AI as merely another “means” of drafting that cannot dilute a lawyer’s responsibility for what is filed.

3. Fletcher v. Experian Info. Sols., Inc., 168 F.4th 231 (5th Cir. 2026)

The court looks beyond the circuit to reinforce a cross-jurisdictional consensus: citation and proposition verification are non-optional and persist irrespective of tools used. Fletcher provides contemporary appellate reinforcement—especially salient because it comes from another federal court of appeals in the same general time period—underscoring that appellate courts are converging on a clear expectation of diligence in AI-era briefing.

4. Sanders v. United States, 176 Fed. Cl. 163 (2025)

Sanders is used to contextualize the “hallucination” problem—cases where counsel cited invented authorities. Critically, the Sixth Circuit uses Sanders to reject a defense-by-distinction: even if the cited cases are real (not invented), misquoting real authorities and misdescribing their holdings remains sanctionable. The opinion thus broadens the cautionary principle from “don’t cite fake cases” to “don’t file unverified AI output, even when it looks plausible.”

5. United States v. Washington, 715 F.3d 975 (6th Cir. 2013) and United States v. Anthony, 280 F.3d 694 (6th Cir. 2002)

These cases appear in the opinion in an unusual posture: not as substantive sentencing precedents applied to decide Farris’s guideline dispute, but as examples of counsel’s misrepresentation.

  • In United States v. Washington, counsel claimed the Sixth Circuit “reversed” a role enhancement on similar facts; in reality, Washington upheld a § 3B1.1 enhancement.
  • In United States v. Anthony, counsel implied vacatur because there was “no evidence” the defendant supervised others; the opinion explains that Anthony vacated narrowly on counting methodology for § 3B1.1(a), and that the defendant actually conceded a supervisory role.

The court’s treatment of Washington and Anthony is pivotal because it demonstrates that the misconduct was not a minor pinpoint-cite error. The briefing error changed the direction of the holdings (affirmance portrayed as reversal; narrow vacatur portrayed as merits-based rejection of supervision). That magnitude supports the court’s remedial choices: replacement counsel, locked briefs, and denial of compensation.

B. Legal Reasoning

1. Technology-neutral ethical baseline

The court’s core move is to treat AI as irrelevant to the existence of the duty, relevant only to the risk profile. It reiterates that lawyers must:

  • Verify citations and propositions before presenting them to a court;
  • Maintain candor and avoid false quotations and misleading characterizations of holdings;
  • Supervise the production of filings, including work delegated to staff and work produced through AI tools.

The opinion frames these obligations through competence and candor and expressly connects them to AI use: if AI increases the probability of plausible-sounding error, then diligence must increase, not decrease.

2. Competence includes understanding the tool’s limitations

The court draws on the competence concept reflected in Model Rules of Pro. Conduct. r. 1.1 cmt. 8, emphasizing the duty to keep up with “relevant technology.” The panel rejects the idea that unfamiliarity with CoCounsel excuses the failure; instead, unfamiliarity is itself a reason not to rely on the tool without guardrails.

The opinion further grounds this in professional guidance, citing A.B.A. Comm. On Ethics & Pro. Resp., Formal Op. 512 (2024) and A.B.A. Task Force on L. & A.I., Addressing the Legal Challenges of AI: Year 2 Report on the Impact of AI on the Practice of Law (2025). These materials are not treated as binding law, but as persuasive statements of what competent practice looks like in an AI environment.

3. Nondelegable responsibility of the attorney of record

A key factual detail is counsel’s reliance on unnamed “staff” to run the AI workflow and his insufficient personal verification. The court treats that reliance as a supervisory failure, citing Model Rules of Pro. Conduct. r. 5.3 and Ky. Sup. Ct. R. 3.130(5.3) (2022). The reasoning is straightforward:

  • Staff assistance is permissible;
  • But the lawyer remains responsible for ensuring accuracy and compliance;
  • AI output (like staff work product) requires lawyer supervision, not rubber-stamping.

4. Harm and institutional integrity justify sanctions and remedial orders

The opinion identifies concrete harms: judicial time spent investigating the irregularities, additional procedural steps (show-cause order, response), and delay in a criminal appeal. The court also highlights a systemic concern: Howe was paid (or would have been paid) through a publicly funded program serving indigent defendants. This aggravates the misconduct because it compromises both efficient adjudication and stewardship of public resources.

5. Tailored remedies: protecting the defendant and the court

The court chooses remedies that address (a) the defendant’s right to meaningful appellate advocacy and (b) the court’s need for accurate briefing:

  • Replacement counsel and re-briefing protect Farris from being bound to defective advocacy.
  • Locking the briefs prevents reliance on contaminated submissions while preserving the record of what occurred.
  • Denial of CJA compensation ensures public funds are not used to pay for work that violated core duties.
  • Disciplinary referrals address broader professional accountability beyond the case.

C. Impact

1. A published Sixth Circuit rule-of-conduct marker for AI-assisted briefing

Because the opinion is recommended for publication, it functions as a durable, citable benchmark within the Sixth Circuit: AI-assisted drafting is permitted, but filing unverified AI-generated text—especially quotations and characterizations of precedent—can trigger concrete sanctions and case-management interventions.

2. “Real case, fake quote” is treated as a serious misrepresentation

A notable contribution of the opinion is its explicit rejection of the notion that only fabricated cases are sanction-worthy. The court treats fabricated quotations and distorted holdings as equally corrosive. Practically, this addresses a common AI failure mode: output that cites real cases but supplies invented language or inaccurate summaries.

3. Heightened expectations for CJA-appointed counsel and public-resource stewardship

The denial of compensation under 18 U.S.C. § 3006A and the court’s emphasis on the “publicly funded program” point to an especially important message for appointed counsel: AI shortcuts that degrade quality can produce not only reputational consequences but also financial consequences (nonpayment) and removal from representation.

4. Procedural ripple effects: more show-cause orders, more verification demands

The opinion models a playbook other panels may adopt: suspicious indicators (like a telltale file name) can justify deeper scrutiny; once triggered, counsel may be required to produce authorities and explain drafting and cite-check processes. This may influence future Sixth Circuit case management when briefing anomalies appear.

IV. Complex Concepts Simplified

  • Criminal Justice Act (CJA), 18 U.S.C. § 3006A. A federal statute providing publicly funded attorneys for defendants who cannot afford counsel. Courts can deny or limit payment when work is not compensable or is improper.
  • U.S.S.G. § 3B1.1(c) leadership-role enhancement. A sentencing guideline that increases a defendant’s offense level by two if the defendant was an organizer, leader, manager, or supervisor in the criminal activity (in cases not meeting larger-scale thresholds in § 3B1.1(a) or (b)).
  • “Show-cause order.” A court order requiring a lawyer (or party) to explain suspected misconduct or deficiencies. Failure to adequately respond can lead to sanctions.
  • AI “hallucination.” A well-known generative AI failure where the system produces plausible-sounding but false information (including invented quotes, citations, or factual claims). The opinion stresses that even when the case citations are real, the quotes and summaries can still be hallucinated.
  • Duty of candor and duty of competence. Professional obligations requiring truthfulness to the tribunal and reasonably skilled, careful representation—both include verifying that cited authorities actually say what the brief claims.
  • Supervision (Model Rule 5.3 / Kentucky counterpart). Lawyers must properly supervise nonlawyer assistants. The opinion extends that practical supervision principle to AI-assisted workflows: delegation does not eliminate attorney responsibility.
  • “Locking” briefs. An administrative step preventing further reliance on filings that the court has determined are unreliable, while maintaining them for transparency and record integrity.

V. Conclusion

United States v. John C. Farris establishes a clear, publication-level Sixth Circuit warning: lawyers may use AI tools, but they must personally ensure the accuracy of quotations, citations, and representations of precedent, and they cannot shift that responsibility to staff or to the tool itself. When counsel files unverified AI-generated briefing that misquotes authorities and misstates holdings, the court may deny CJA compensation, remove counsel, order re-briefing, lock defective filings, and refer the lawyer for discipline. The opinion thus reinforces that technological innovation does not relax professional duties—if anything, it demands more rigorous verification.