USA v. Farris: Sixth Circuit Removes Attorney for CoCounsel AI Hallucinations | Advocate Prakhar

Case at a Glance

Full Citation	United States v. Farris, No. 25-5623 (6th Cir. April 3, 2026) [Recommended for Publication]
Court	U.S. Court of Appeals for the Sixth Circuit
Date	April 3, 2026
Judges	Clay, Gibbons, and Hermandorfer, Circuit Judges
Category	AI Hallucination — CJA Criminal Defense; Attorney Removal; Bar Referral
Jurisdiction	USA — Federal (Sixth Circuit)
AI Tool	CoCounsel (Westlaw / Thomson Reuters AI platform)
Outcome	Attorney removed; CJA compensation denied; referral to Chief Judge and Kentucky Bar; replacement counsel appointed; briefs locked

Background

John C. Farris pleaded guilty to drug-trafficking charges and was sentenced to include a two-level leadership enhancement under U.S.S.G. § 3B1.1(c). His court-appointed attorney under the Criminal Justice Act, Steven N. Howe, appealed the enhancement. Howe filed a principal brief and a reply brief on Farris’s behalf. The Sixth Circuit first noticed something suspicious before completing its substantive review: the file name of the principal brief was “CoCounsel Skill Results” — revealing that the brief was generated directly from Westlaw’s AI platform without independent composition.

The AI Issue

On substantive review the court found three fabricated quotations: (1) a quotation attributed to U.S.S.G. § 3B1.1, commentary n.1 that does not appear in the Guidelines; (2) a quotation from United States v. Washington, 715 F.3d 975, about decision-making authority that does not appear in that case; (3) a quotation from United States v. Anthony about evidence of direction that similarly does not exist. The first quotation’s fabrication is attributed to real Guidelines text that does not contain the bracketed material; the latter two appear to be AI-invented quotations inserted into real citation wrappers. Howe admitted using CoCounsel, acknowledged the fabrications, and conceded he failed to adequately review and verify the AI output.

What the Court Decided

• Howe’s reliance on “staff” to supervise AI-generated content did not satisfy his obligations as attorney of record — supervision of AI must be done by the attorney, not delegated to non-attorneys. [non-delegable supervisory duty]
• That the citations referenced real cases (not wholly fictitious ones) did not absolve Howe — fabricated quotations within real citations still violate the duty of candour. [partial accuracy no defence]
• Howe was removed from further representation of Farris; replacement counsel appointed under CJA. [attorney removal]
• Howe denied all CJA compensation for the appeal — public funds will not pay for AI-generated misconduct. [compensation forfeiture]
• Copies of the opinion referred to the Chief Judge of the Sixth Circuit for disciplinary proceedings, the E.D. Kentucky District Court, and the Kentucky Bar Association. [multi-forum bar referral]
• Howe’s briefs locked; briefing schedule reset upon appointment of replacement counsel, further delaying Farris’s appeal. [harm to indigent defendant]

“Attorneys who choose to use artificial-intelligence tools must do so in a manner consistent with their ethical obligations… attorneys who rely on artificial intelligence must remain diligent in supervising their work product and carefully examine the accuracy of every citation they present to this Court.”

— Sixth Circuit Court of Appeals (Clay, Gibbons, Hermandorfer, JJ.), April 3, 2026

The India Angle

Indian Law Equivalent

India’s system of Legal Services Authorities under the Legal Services Authorities Act 1987 provides free legal aid to indigent accused in criminal matters. Advocates appearing on legal aid panels under the National Legal Services Authority have the same professional obligations as privately retained counsel. An advocate who uses AI without verification in a legal aid matter fails not only the court but the constitutional guarantee of fair trial under Article 21 of the Constitution — a dimension absent in civil AI hallucination cases.

Bar Council Rules

Rule 44 of the BCI Rules requires advocates to attend to their cases diligently and with reasonable competence. For legal aid advocates, the Supreme Court has held in Hussainara Khatoon v. State of Bihar that the right to legal representation must be meaningful, not merely formal. An AI-hallucination brief in a criminal appeal fails the “meaningful representation” standard and could be grounds for retrial or fresh appellate proceedings at the state’s cost.

Practical Advice for Indian Advocates

• Legal aid panel advocates must understand that the Farris standard applies equally to them — public funding does not lower the verification duty; it raises the stakes because an unverified AI brief may delay justice for an imprisoned client.
• In sessions court appeals and High Court criminal revisions, every citation to the Sentencing Guidelines equivalent (Indian Penal Code tariff provisions) must be cross-checked against the actual statutory text and Supreme Court sentencing decisions.
• The discovery that a file is named “CoCounsel Skill Results” or similar AI-generated nomenclature is a red flag that courts will increasingly look for — rename and verify all AI-generated drafts before filing.

Quick Takeaways

• A suspicious file name can trigger judicial scrutiny before the court even reads the document.
• CJA compensation forfeiture means the attorney bears the full financial cost of AI misconduct — there is no public subsidy for irresponsible AI use.
• Fabricated quotes in real cases are just as sanctionable as wholly invented case citations.

---

Deep Dive: AI in Criminal Defense — Constitutional Stakes and the CoCounsel Moment

United States v. Farris represents the highest-stakes context in which AI hallucination has so far been adjudicated: a criminal sentencing appeal in which a court-appointed attorney, funded by public money to represent an indigent defendant, used an AI tool without adequate verification. The constitutional dimension here is not theoretical. Under the Sixth Amendment’s guarantee of effective assistance of counsel, an attorney who submits fabricated legal authority to a court has arguably failed to provide the minimum standard of representation — a claim that Farris himself may eventually raise in a § 2255 motion if his delayed appeal is unsuccessful.

The Sixth Circuit’s handling of the case is notable for several reasons. First, the court detected the AI use before completing substantive review — the file name “CoCounsel Skill Results” was the tell. This suggests that courts are developing new front-line detection methods beyond simply catching misquoted cases during legal research. File metadata, document naming conventions, stylistic inconsistencies, and citation pattern analysis are all becoming tools in the judicial review process. Attorneys using AI should understand that the pipeline from prompt to filing may leave digital fingerprints visible to careful judicial review.

Second, the court’s treatment of the CJA compensation issue is significant. Under 18 U.S.C. § 3006A, court-appointed attorneys are compensated from public funds for time spent on representation. The Sixth Circuit’s denial of all CJA compensation for the appeal means that Howe spent significant time on a brief he cannot bill for — a financial deterrent that may be more effective than a symbolic monetary sanction. For a solo practitioner or small firm attorney working under CJA rates, losing the compensation for an entire appeal is a substantial economic consequence.

For Indian legal aid advocates, the Farris case should be read alongside the Supreme Court’s consistent emphasis, since Suk Das v. Union Territory of Arunachal Pradesh (1986), that legal aid must be meaningful and effective. If an Indian legal aid lawyer uses AI to draft a criminal revision petition and that petition is dismissed for fabricated citations, the prisoner-client has suffered a concrete harm — wasted appeal opportunity, continued incarceration — that could form the basis of a complaint before the State Legal Services Authority and potentially a contempt action against the advocate. The duty of competence in criminal representation is not lessened by the availability of AI tools; it is heightened.