Gentry v. Thompson: New Orleans City Attorneys Sanctioned $1,250 for Nine ChatGPT-Hallucinated Citations | Advocate Prakhar

Case at a Glance

Background

Daniel Gentry filed a civil rights lawsuit in June 2025 against the City of New Orleans, then-Mayor LaToya Cantrell, the New Orleans Police Department, and individual NOPD officers for violations of his constitutional rights and tortious conduct. In January 2026, the City Defendants filed a Rule 12(b)(6) motion to dismiss with a supporting memorandum signed by four city attorneys: Jalen Harris (the drafter), James Roquemore (his supervisor), Corwin St. Raymond (Chief Deputy City Attorney), and Charline Gipson (newly appointed City Attorney).

When the court reviewed the supporting memorandum, it realised that nine of the case citations were non-existent — apparent ChatGPT hallucinations. On March 5, 2026, it issued a Show Cause Order. At the March 19 hearing, Harris admitted he had started with Westlaw research but switched to ChatGPT, and that he neither checked the ChatGPT-provided citations nor read the nine hallucinated cases he submitted. Notably, the hallucinated citations were formatted in bullet points — an unusual style that should have signalled the need for closer examination by anyone reviewing the brief.

The AI Issue

The case squarely presents the question of how responsibility distributes when AI-generated research flows through a hierarchical law office. Harris drafted and used ChatGPT; Roquemore supervised and reviewed, making substantial revisions to other sections but failing to examine the bullet-pointed citations; St. Raymond and Gipson were on the signature block. The court’s analysis was calibrated: Harris violated Rule 11 by not checking citations generated by an unreliable tool. Roquemore, with 30 years of experience, bore greater responsibility because his supervisory role obligated him to ensure that every cited case had been read — especially when the citations were formatted in an unusual way that should have triggered inquiry.

What the Court Decided

• Jalen Harris personally sanctioned $250 for using ChatGPT without verifying the hallucinated citations, in violation of Rule 11.
• James Roquemore personally sanctioned $1,000 for failing in his supervisory duty to verify citations, particularly given his 30 years of experience and the unusual formatting of the hallucinated entries.
• No sanction imposed on Corwin St. Raymond, but the order serves as a formal admonishment regarding AI misuse in his department.
• City Attorney Gipson not sanctioned — she had been sworn in only three days before the filing.
• The City Attorney’s office was required to develop a written AI use policy, and Harris and Roquemore were directed to attend a Thomson Reuters seminar on AI in legal practice.
• The court also found that Defendants had failed to serve the pro se plaintiff with pending motions — an additional procedural failure unrelated to AI but corrected at the hearing.

“Rule 11 implicitly requires that attorneys read the cases they cite in court filings to determine whether their arguments are supported by law. Using ChatGPT and failing to check the case citations it provides constitute not only a lapse in professional judgment, but also a violation of Federal Rule of Civil Procedure 11.”

— Judge Carl J. Barbier, E.D. Louisiana, March 20, 2026

The India Angle

Indian Law Equivalent

Rule 11 of the Federal Rules of Civil Procedure has its closest Indian analogue in Order VI Rule 15 of the Code of Civil Procedure (CPC), 1908, which requires a pleading to be verified by the party or a person acquainted with the facts, and in the inherent powers of courts under Section 151 CPC to prevent abuse of process. In the context of frivolous or misleading submissions to Indian courts, the Supreme Court has consistently exercised its contempt jurisdiction under the Contempt of Courts Act, 1971. A lawyer who supervises a junior’s work and signs off on AI-hallucinated citations could face contempt under Section 2(b) (civil contempt for wilful disobedience of court norms) or Section 2(c)(i) (criminal contempt for scandalising the court).

Bar Council Rules

BCI Rule 36 requires advocates to communicate honestly with clients, but the supervisory dimension is less explicitly codified in India than in the US. However, the principle is well-established through case law: a senior advocate who endorses a junior’s work as fit for submission takes on responsibility for its accuracy. The duty of candour under Rules 49 and 52 falls on every signatory of a court submission. Indian law firms and chambers should consider implementing internal AI verification protocols, particularly since the BCI has not yet issued formal guidance on AI use in legal practice.

Practical Advice for Indian Advocates

• Senior advocates who supervise juniors must implement a citation-verification step that is independent of the drafting process — a junior who used AI to draft is not a reliable checker of their own AI’s citations.
• Unusual formatting in a draft brief (such as bullet-pointed case citations) is a red flag that the citations may be AI-generated and unverified — treat it as a mandatory review trigger.
• Developing a firm-level AI use policy — specifying permissible uses, verification requirements, and disclosure obligations — reduces professional and institutional risk and is likely to become a regulatory expectation as courts and Bar Councils develop AI guidelines.

Quick Takeaways

• Supervisors who sign briefs bear Rule 11 responsibility proportional to their experience and supervisory authority — ignorance of a junior’s AI use is not a shield.
• Nine hallucinated citations is an extraordinary failure — but the real problem is zero verification by anyone in the review chain.
• Courts are now ordering affirmative steps: AI-use policies, CLE compliance, and formal admonishments for government law offices that fail to prevent citation hallucinations.

Deep Dive: Supervisory Responsibility in the AI Era — The Roquemore Standard

The most legally significant aspect of Gentry v. Thompson is not the sanction of the junior attorney — that was expected — but the sanctioning of the supervising attorney at a higher level. This establishes what we might call the “Roquemore Standard”: a supervising attorney who reviews a brief before filing but fails to check citations for reliability, particularly when those citations exhibit unusual formatting or other red flags, violates Rule 11 and bears heightened responsibility commensurate with their experience.

This standard has profound implications for law firm management in the AI era. In traditional legal practice, a supervisor’s review of a brief was primarily substantive — did the argument hold together, was the writing clear, did the strategy make sense? Citation verification was largely left to the drafter, on the assumption that the drafter had done primary research. AI fundamentally changes that assumption. A brief drafted with AI assistance may look substantively coherent and be stylistically polished while containing citations the drafter never read. A supervisor who reviews the substance without independently verifying citations is relying on a check that the drafter may not have performed.

The solution is structural: law firms need to separate the verification function from the drafting function. The person who verifies citations should not be the same person who generated them using AI. A senior associate, a paralegal trained in citation verification, or a dedicated research role should independently confirm each citation in any brief before it is filed. This is particularly important because AI-generated citations often look plausible — they have realistic parties, realistic reporters, realistic years. Only by actually opening the cited case in Westlaw or Lexis can the fabrication be detected.

The Gentry case also highlights the institutional response courts expect. The City Attorney’s office was told to develop a written AI policy and send attorneys to CLE on AI use. Courts are now treating AI-hallucination sanction orders not as isolated individual failures but as institutional failures requiring systemic correction. This approach will likely spread: future courts in similar cases may demand proof of institutional AI-governance measures as part of the sanction response.