Smith v. Verra Mobility: ChatGPT Hallucinations Trigger Client Notification Order | Advocate Prakhar

Case at a Glance

Full Citation	Smith v. Verra Mobility Corp. et al., Case No. 8:25-cv-2300-KKM-NHA, 2026 WL 1197241 (M.D. Fla. May 1, 2026)
Court	U.S. District Court, Middle District of Florida (Tampa)
Date	May 1, 2026
Judge	Kathryn Kimball Mizelle, U.S. District Judge
Category	AI Hallucination — ChatGPT; Minimal Sanction; Client Notification
Jurisdiction	USA — Federal (Florida)
AI Tool	ChatGPT (explicitly named)
Outcome	Non-monetary sanction: attorney ordered to provide client with copies of show-cause order, response, and this order

Background

Richard Paul Smith filed a class action suit against Verra Mobility Corporation and other defendants including the Polk County School Board in the Middle District of Florida. The Polk County School Board’s lead counsel, attorney Jeffrey Sullivan, filed a motion to dismiss that the court later found included several misrepresentations of caselaw and citations to non-existent cases. On April 28, 2026, Judge Mizelle issued a show-cause order directing Sullivan to explain why he should not be sanctioned or referred to the Florida Bar. Sullivan responded on May 1, 2026.

The AI Issue

Sullivan’s response admitted that he had used artificial intelligence tools, including ChatGPT, to research and draft the School Board’s motion to dismiss. He acknowledged that he did not “complete an independent verification of each citation before filing” and did not realize the cited authorities were inaccurate or unsupported. Sullivan accepted full personal responsibility, did not blame the School Board, and outlined a plan to ensure all future cited authorities are “personally reviewed, accurately described, and independently verified before filing.” This candid and complete response was central to the court’s decision to impose only minimal sanctions.

What the Court Decided

• The use of AI does not reduce or alter an attorney’s duties to the court under Rule 11 or to the client. [non-delegable professional duty]
• Sullivan violated Rule 11(b)(2) by failing to verify multiple cited authorities, constituting negligent (not intentional) misconduct. [objective reasonableness standard]
• Immediate and unqualified acceptance of responsibility is a significant mitigating factor justifying minimal sanctions. [mitigation doctrine]
• The “most fitting sanction” was client notification — Sullivan was ordered to provide the School Board with copies of the show-cause order, his response, and this ruling. [non-monetary sanction]
• No referral to the Florida Bar; no monetary penalty; the client notification serves as the deterrent. [proportionality in sanctions]

“The use of artificial intelligence does not alter or minimize his duties to the Court under Rule 11, nor the separate duties owed to his client.”

— Judge Kathryn Kimball Mizelle, M.D. Fla., May 1, 2026

The India Angle

Indian Law Equivalent

In India, the attorney-client relationship is governed by the Advocates Act 1961 and the BCI Rules. Rule 19 of the BCI Rules requires advocates to inform clients of all significant developments in their matters. An advocate who submits AI-hallucinated citations in a client’s case arguably violates this duty of disclosure — the client has a right to know that their legal submissions contained fabricated authority, which may have prejudiced their case.

Bar Council Rules

Rule 24 of the BCI Rules prohibits advocates from suppressing any material fact from the client. The Smith v. Verra approach — mandating that the attorney inform the client of the AI-hallucination incident — is consistent with Rule 24’s spirit. Indian bar councils should consider issuing guidance requiring advocates who discover post-filing AI hallucination errors to promptly disclose the same to their clients, mirroring the client notification sanction model.

Practical Advice for Indian Advocates

• When using ChatGPT or any LLM for research, adopt a zero-trust verification policy: every single citation must be independently confirmed against an authoritative database before including it in any court document.
• Build a “ChatGPT used — verified” notation into your file management system so that each citation’s verification status is tracked and can be demonstrated if a court questions accuracy.
• If you discover a hallucinated citation after filing, immediately inform your client, file a correction with the court, and notify opposing counsel — the Smith v. Verra approach shows that candour is the most effective risk mitigation tool available.

Quick Takeaways

• ChatGPT hallucinations in court filings violate Rule 11 even when the attorney acted without intent to deceive.
• Complete and immediate acceptance of responsibility can reduce sanctions to a client notification order.
• Client disclosure is itself a meaningful sanction — professional reputation with clients is a key deterrent.

---

Deep Dive: The Spectrum of AI Sanctions — From Warning to $5,000 and Beyond

Smith v. Verra Mobility sits at the lenient end of the growing spectrum of AI hallucination sanctions, and examining where it falls on that spectrum — and why — reveals the factors courts weigh in calibrating their response. At the harsh end: Mata v. Avianca (S.D.N.Y. 2023) imposed $5,000 in monetary sanctions; Johnson v. Dunn (N.D. Ala. 2025) imposed public reprimands and financial penalties; United States v. Farris (6th Cir. 2026) stripped an attorney of CJA compensation and referred him for disciplinary proceedings. Smith v. Verra imposed only a client notification order. The difference is not in the nature of the violation but in the attorney’s response.

Judge Mizelle’s analysis identifies four key mitigating factors: (1) Sullivan’s immediate and unqualified acceptance of responsibility; (2) the absence of any attempt to shift blame to the client or to colleagues; (3) a concrete and articulated plan for preventing future violations; and (4) the recognition that this appeared to be the first such incident in Sullivan’s practice. These factors collectively justify the minimum sanction necessary to deter repetition — client notification — rather than a monetary penalty or bar referral.

The client notification sanction is clever in its design. It operates on the professional reputation harm channel rather than the financial channel. An attorney ordered to hand-deliver to a government school board client a judge’s ruling that the attorney used ChatGPT irresponsibly and filed fabricated citations faces a real consequence — the loss of the client’s trust, potential fee disputes, and reputational damage that far exceeds any moderate monetary fine. This sanction also creates an accountability loop: the client can then make informed decisions about whether to continue the representation and whether to independently verify future filings.

For Indian government advocates and panel lawyers — who represent state boards, PSUs, and government departments — the client notification model is particularly instructive. Government clients are accountable to legislatures, audit bodies, and the public. A government advocate who uses AI without verification and files fabricated citations is not just exposing himself to court sanction; he is potentially creating a compliance failure for the government body he represents. Indian government departments should begin requiring their legal panels to certify, as part of engagement terms, that all citations in pleadings have been independently verified against authoritative databases — building the verification requirement into the contractual framework before it becomes a sanction-driven obligation.