The 11th Circuit referred Anthony Sabatini for replacing eight hallucinated cases with eight more
On July 10, 2026, the U.S. Court of Appeals for the Eleventh Circuit published Estate of Lane Caviness v. Atlas Air (No. 24-11033), reprimanding Florida lawyer and Lake County Commissioner Anthony Sabatini for briefs 'replete with fake and hallucinated citations.' His opening brief relied on at least eight nonexistent cases, including one supposedly decided by the Eleventh Circuit itself. After opposing counsel flagged them, Sabatini filed an untimely reply withdrawing eight cases, and none matched the original eight, and all eight replacements were also fabricated. Judge Britt Grant wrote that 'whatever the merits of artificial intelligence, it is no substitute for actual intelligence' and referred Sabatini to the court's Committee on Lawyer Qualifications and Conduct.
Records by entity: Sabatini Law Firm
A federal appeals court called out a lawyer by name in a published opinion for AI-fabricated case law, then watched him withdraw the wrong eight cases, all fake too. He now faces a disciplinary referral.
Whatever the merits of artificial intelligence, it is no substitute for actual intelligence.
Key facts
- What
- On July 10, 2026, the U.S.
- Incident date
- Jul 10, 2026
- Who
- Sabatini Law Firm
- Failure mode
- Hallucination
- AI surface
- Chatbot
- Severity
- Medium
What happened
Sabatini represented aviation workers suing Atlas Air over Covid-era vaccine, testing, and mask policies. The Eleventh Circuit affirmed dismissal, calling the discrimination claims 'wildly implausible,' then devoted Section V of the opinion to the filings themselves. The opening brief cited at least eight nonexistent decisions, listed by the court in a footnote, fake citations and all. Confronted, Sabatini acknowledged 'erroneous or unverifiable' citations and sought to withdraw eight authorities; the panel found none matched the original eight and all eight replacements were also hallucinated, with one phantom Delta Air Lines decision migrating between courts across the two filings. The unanimous panel held that completely outsourcing legal work to AI 'is not competent,' violated duties to clients and court, taxed appellate costs against Sabatini's clients, and referred him through the Chief Judge to the court's lawyer-conduct committee. It is among the sharpest published AI-sanctions opinions from any federal appeals court to date.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
A generative tool produced fluent, confidently formatted legal authority with no grounding in any reporter, and the lawyer's cleanup pass ran the same failure loop a second time: asked to identify the fabrications, the workflow generated a new set of plausible-looking citations rather than retrieving the actual contents of the original brief. The court's footnote observation applies mechanically: these systems echo what the user wants to hear, and a lawyer whose verification step is the same tool that fabricated the material has no verification step at all.
What it cost
Sources
- Court FilingEstate of Lane Caviness v. Atlas Air, Inc., No. 24-11033 (11th Cir. July 10, 2026)media.ca11.uscourts.gov
- PressEx-Florida Congress Candidate Slammed by Court Over AI in Briefsnews.bloomberglaw.com
- PressLawyer's use of 'fake and hallucinated' citations gets reprimand from US appeals courtabajournal.com
Cite this entry
https://failureindex.ai/failures/sabatini-eleventh-circuit-hallucinated-citations-referralAI Failure Index. "The 11th Circuit referred Anthony Sabatini for replacing eight hallucinated cases with eight more" (FI-0728). Realm Labs. https://failureindex.ai/failures/sabatini-eleventh-circuit-hallucinated-citations-referral (indexed Jul 17, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0728. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
Prism reads the internal signature of a model asserting citations it has no grounded support for, the hallucination fingerprint, before the text lands in a filing. OmniGuard holds legal-citation outputs at the boundary until each authority verifies against a real citator, blocking fabricated case law from ever reaching a signature line.