Important date caveat

The incident_date is July 10, 2026 because that is the documented date of the Eleventh Circuit’s opinion and the resulting public response. It is not the date the briefs were filed or AI was used; the audited sources do not establish either of those dates.

What happened

On July 10, 2026, the U.S. Court of Appeals for the Eleventh Circuit affirmed the district court’s dismissal of a complaint brought by aviation employees challenging pandemic-era workplace policies. The court then devoted the end of its published opinion to the conduct of plaintiffs’ counsel Anthony F. Sabatini.

According to the opinion, the opening brief relied on at least eight nonexistent cases. After the defendants flagged the problem, counsel submitted an untimely proposed reply acknowledging eight “erroneous or unverifiable” citations and seeking to withdraw reliance on them.

Those were not the same eight authorities the defendants had identified in the opening brief. The court found that the second list was hallucinated too.

What crossed the boundary

The court found that AI-assisted legal work entered signed appellate filings without verification that the cited cases existed. After the first problem was identified, counsel sought withdrawal through the untimely proposed reply; the audited record does not establish that the court accepted it as a correction.

The court described the work as outsourced to an AI algorithm and said counsel remained responsible for the machine-produced work he signed. That finding establishes the reliance decision without establishing who operated the system or how the work moved from the system into the filings.

What went sideways

The formal disposition was separate from the citation problem. The Eleventh Circuit affirmed dismissal of the third amended complaint on mixed grounds: lack of personal jurisdiction as to Flight Services International and failure to state a claim as to Atlas Air. It did not dismiss the appeal, and the disposition was not entirely a merits ruling.

The citation failures drew public judicial criticism and an announced disciplinary-process step. The court said that, through the Chief Judge and in a separate order, it would refer the matter to its Committee on Lawyer Qualifications and Conduct. The audited sources do not establish that the separate order issued, that the referral occurred, or that any sanction, final disciplinary finding, or completed committee proceeding followed.

What the evidence supports

The official opinion directly documents the court’s findings about unverified AI use, nonexistent authorities in the opening brief, the proposed withdrawal, the second hallucinated set, and the announced future referral. It also documents that the complaint was dismissed on jurisdictional and pleading grounds, not because AI was used.

Bloomberg Law independently reported on the decision on July 10 and reported no immediate response to its requests for comment from Sabatini and attorneys for both parties. Its central account derives from the opinion; the reported nonresponse is neither an admission nor separate proof of the court’s findings.

The “Confirmed” confidence label applies only to this court-documented core. It does not confirm a completed referral, a sanction, the date of the underlying conduct, or details of the AI workflow.

What remains unknown

The audited sources do not give the filing dates of the opening brief or proposed reply, or the date AI was used. They do not identify the AI product, model, prompts, original output or transcript, logs, operator, or exact workflow. Without those records, the article cannot reconstruct how the citations were generated or who directly used the system.

The practical lesson

A fluent citation is not a verified authority. In a profession built around checkable sources, opening the cited case is a basic verification step, and signing a filing leaves responsibility with the person who submits it.