Hessert v. Hessert: Florida Appellate Court Catalogues AI-Hallucination Crisis, Finds 9 of 13 Citations Defective | Advocate Prakhar

Case at a Glance

Full Citation	Melissa Elizabeth Harrell Hessert v. Gary Bell Hessert and Megan Wieland-Pulayya, 2026 WL 785016 (Fla. 6th DCA Mar. 20, 2026)
Court	District Court of Appeal of Florida, Sixth District
Date	March 20, 2026
AI Tool / Issue	Pro se certiorari petition: 5 of 13 cited cases non-existent; 4 of 13 cited for propositions not found in the opinions; only 4 valid citations
Outcome	Petition denied without further discussion on merits; court issues comprehensive warning on AI use; obligation on opposing counsel to flag AI errors established

Background

Melissa Hessert, proceeding pro se, filed a petition for writ of certiorari in the Florida 6th DCA challenging a circuit court order in a family law matter (post-dissolution proceedings, Case No. 2020-DR-002544). The petition itself was denied without further comment on its merits. The court wrote separately to address the petition’s citation problems, which it characterised as bearing all the hallmarks of AI-generated content that was not verified.

The numbers are stark: of 13 cases cited in the petition, only 4 both exist and are cited for legal propositions that the cited cases actually represent. Five cases do not exist at all. Four cases exist but are cited for legal propositions that the cited cases do not represent. A 69% error rate in a legal brief — nearly three-quarters of all citations defective — represents an extraordinary failure of professional diligence.

The AI Issue

The court used the Hessert ruling to deliver its most comprehensive statement yet on AI hallucinations in Florida appellate proceedings. It catalogued the national scope of the problem, surveyed Florida’s own growing line of AI-citation cases across multiple districts, and articulated the dual obligation that the ruling imposes: on filers, to verify citations before filing; and on opposing counsel, to point out AI errors in responsive filings or by motion. This second obligation — on opposing counsel — is notable as a practical enforcement mechanism that does not rely on courts independently detecting the errors.

What the Court Decided

• Petition for writ of certiorari denied without further discussion on the underlying merits.
• Court wrote separately to address the AI-citation problem.
• Both pro se litigants and attorneys required to: (1) cite-check cases to ensure they exist; and (2) cite-check cases to ensure they support the propositions asserted — as a condition of signing any appellate filing.
• Opposing counsel who discover AI-fabricated citations in the other party’s filings should point out such errors in responsive filings or via motion.
• Florida appellate rules cited: Florida Rule of Appellate Procedure 9.045(d); Florida Rule of General Practice and Judicial Administration 2.515(d)(2).
• Attorneys reminded of duties under Florida Bar Rules 4-1.1 (competence), 4-1.3 (diligence), and 4-3.3 (candour toward tribunal).

“[I]f and when any such erroneous filings [containing nonexistent cases or inaccurate legal propositions] are made … filers on the opposite side of the case should point out such errors … in their responsive filings or via motion.”

— Judge Pratt, Hessert v. Hessert, Fla. 6th DCA, March 20, 2026

The India Angle

Indian Law Equivalent

India’s equivalent professional duties are embedded in the Bar Council of India Rules, particularly Rule 49 (no misstatement of law) and Rule 52 (accurate citation). Indian appellate courts — from High Courts to the Supreme Court — rely on the Bar’s duty of candour and on the principles of audi alteram partem and fairness. A petition that cites 13 cases of which 9 are defective would likely be treated as an abuse of process in an Indian appellate court, and the court’s inherent jurisdiction under Articles 136 and 142 of the Constitution would support both the dismissal and a reference to the Bar Council.

Bar Council Rules

The Hessert court’s reference to dual obligations — on filers to verify, on opposing counsel to flag errors — has a parallel in India’s adversarial system. BCI Rule 24 requires advocates to be fair and honest to the court; Rule 30 requires advocates not to allow the court to be misled by any statement of fact or law. Where opposing counsel discovers AI-hallucinated citations in an adversary’s brief, the duty of fairness to the court arguably requires them to bring this to the court’s attention — though practice varies and this duty is less explicitly stated in India than in the Florida ruling.

Practical Advice for Indian Advocates

• A 69% citation error rate is not a minor oversight — it is a systemic failure of the drafting and review process. Any workflow that produces this level of error requires complete overhaul before any further filings are made.
• When opposing counsel submits a brief with citations you cannot verify, point this out in your responsive brief — the Hessert ruling establishes this as an obligation in Florida and it reflects general principles of candour that are applicable in India.
• Two-step verification (existence check + proposition accuracy check) is the minimum standard articulated by multiple courts — build this as a non-negotiable final step in every appellate brief preparation process.

Quick Takeaways

• A 69% citation defect rate (9 of 13 citations defective) in a single brief represents the most severe documented example of AI-hallucination magnitude in reported case law.
• Florida courts now impose a dual obligation: filers must verify before filing; opposing counsel must flag errors when discovered — making AI-hallucination detection a shared responsibility of the adversarial process.
• The two categories of defective citation (non-existent case vs. existing case misrepresented for wrong proposition) require different verification steps and are both caught only by the two-step verify-and-read approach.

---

Deep Dive: The Anatomy of a 69% Error Rate — Understanding AI’s Two Failure Modes in Legal Citation

The quantification in Hessert v. Hessert — five non-existent cases, four real cases misrepresented — reveals the two distinct failure modes of AI in legal citation generation, and why they require different responses. Understanding these modes is essential for advocates who want to implement effective verification workflows.

Failure Mode 1: Fully fabricated citations. The AI generates a party name, a reporter citation, a year, and a jurisdiction — all plausible-sounding but corresponding to no actual case. This is the failure mode that receives most attention because it is the most dramatic, but it is also the most easily caught: a basic search on Westlaw, Lexis, Google Scholar, or any authenticated legal database returns no result. Five minutes of checking is all that is needed to detect every instance of this failure mode.

Failure Mode 2: Real case, wrong proposition. The AI correctly identifies a real case that exists in the jurisdiction, often one that is broadly related to the subject matter, but attributes to it a legal proposition that is not in the case. This failure mode is significantly harder to catch because the existence check passes — the case is there. Only reading the relevant sections of the opinion reveals that the stated proposition is absent. This is the more intellectually challenging verification task, but it is also the more legally dangerous one, because opposing counsel who does a quick existence check may miss the error entirely.

The two-step verification protocol that Florida courts (and increasingly all common-law courts) are demanding addresses both failure modes: step one catches fully fabricated citations; step two catches real cases misrepresented. The Hessert ruling is important precisely because it names both categories explicitly, helping practitioners understand what verification must actually accomplish.

For law firms and advocates in India building AI governance policies, the two-failure-mode framework provides a practical structure for verification checklists. Every citation should be checked for (1) existence — does this case appear in an authenticated database at this citation? — and (2) proposition accuracy — does the relevant section of this case contain the legal proposition for which it is being cited? Both checks are necessary; neither is sufficient alone. The Hessert data — five failures of Mode 1, four failures of Mode 2, in a single 13-citation brief — shows that AI fails both modes at high rates in legal contexts.