AI-Invented Case Law — From Lawyers' Briefs to Judges' Orders

On 22 October 2026 the US Court of Appeals for the Fifth Circuit took a case away from a federal district judge in Mississippi. In Jackson Federation of Teachers v. Fitch (No. 25-60496), a panel of Judges Smith, Haynes and Engelhardt vacated Judge Henry Wingate's preliminary injunction and directed the chief judge of the Southern District of Mississippi to reassign the case to a different judge. One reason was that the judge's orders had named parties who were never in the case and cited a case that does not exist. The judge had said a law clerk used AI to draft one of them.

Plausible but false output from generative AI — hallucination — first reached court records through lawyers' filings and then reached the decisions of courts themselves. This page sets out three turning points.

2023: lawyers file ChatGPT's case law

The first widely known case was a suit against the airline Avianca by a passenger who said a metal serving cart struck his knee on a flight to New York. On 1 March 2023 the plaintiff's lawyers filed a submission in the Southern District of New York that cited judicial opinions that did not exist. Steven A. Schwartz researched and wrote it; Peter LoDuca of the same firm signed and filed it.

In his Opinion and Order on Sanctions of 22 June 2023, Judge P. Kevin Castel found that the two lawyers and their firm, Levidow, Levidow & Oberman, had submitted non-existent judicial opinions with fake quotes and citations created by the AI tool ChatGPT, and had continued to stand by them after court orders called their existence into question. He opened by writing that "there is nothing inherently improper about using a reliable artificial intelligence tool for assistance," while stressing that existing rules give lawyers a gatekeeping role over the accuracy of what they file.

The sanctions: a US$5,000 penalty, paid jointly and severally into the court's registry; letters to the client enclosing the order; and letters to each real judge falsely named as the author of the six fake opinions — "Varghese", "Shaboon" and the rest — enclosing the order and the hearing transcript. The judge declined to order an apology, reasoning that a compelled apology is not a sincere one.

July 2026: India's Supreme Court sets aside a tribunal's order

In the next turning point the fake precedents appeared in the decision-maker's own text. On 28 August 2024 the National Company Law Tribunal (NCLT) in Mumbai admitted an insolvency application by Jammu and Kashmir Bank against Essel Infraprojects. On 11 September 2025 the National Company Law Appellate Tribunal (NCLAT) dismissed the appeal.

When Pooja Ramesh Singh, a suspended director of the company, took the case to the Supreme Court, the precedents the NCLT had relied on became the issue. According to Bar & Bench, the Court's own verification found that some citations were entirely fictitious while others carried fabricated extracts. The bank stated in an affidavit that its counsel had not cited them and that the tribunal had found them through its own research.

On 2 July 2026 Justices PS Narasimha and Alok Aradhe set aside the NCLT and NCLAT rulings and restored the application to the NCLT. The judgment called citing AI-generated precedents without verification misconduct by an advocate, and a judge's reliance on them a serious lapse: "It is necessary for Courts to adopt a zero-tolerance mode for producing, citing or using AI-generated precedents without verification." Such a decision, it said, is no decision in the eyes of the law. It directed the Bar Council of India to form a committee and lay down disciplinary standards.

The same judgment affirmed the Court's resolve to use AI in aid of adjudication while keeping total control over adjudication with a human in the loop. It described the use of AI-generated fake material as precedent as "like the release of methyl isocyanate in the province of law and justice."

October 2026: a US appeals court replaces the judge

In April 2025 Mississippi enacted H.B. 1193, which among other things bars public educational institutions from mandating diversity, equity and inclusion (DEI) programmes and trainings. In June, teachers' groups, students and parents brought a pre-enforcement facial challenge.

On 20 July 2025 Judge Wingate issued a temporary restraining order (TRO) blocking the law. According to the Fifth Circuit, it named as plaintiffs eleven individuals and organisations that had never been parties, named as defendants two officials who had not been sued, referred to purported declarations from four people who do not appear in the record, and misquoted the statute. When the state moved for clarification on 22 July, the judge replaced the order without explanation and backdated the new version to 20 July. He refused to restore the original to the docket, calling the errors "clerical."

In October 2025 Senator Chuck Grassley, chairman of the Senate Judiciary Committee, asked whether AI had been used. The judge replied that a law clerk had used AI to help draft an earlier version that was docketed by mistake; according to Mississippi Today and the ABA Journal, his letter identified the tool as Perplexity. In the same round of letters a federal judge in New Jersey said a law-school intern had used ChatGPT, and, Mississippi Today reported, the director of the Administrative Office of the US Courts said he had set up a task force to recommend policies on AI use.

Meanwhile the judge issued a preliminary injunction on 18 August, and the state appealed. At oral argument it emerged that even the corrected TRO cited a non-existent Fourth Circuit case, "Cousins v. School Bd. of City of Norfolk" — and the preliminary injunction incorporated the TRO's analysis by reference.

Judge Jerry E. Smith's opinion was blunt. It called the district-court proceedings "a botch job, to say the least," and said that once confronted with the falsehoods, the judge insisted they had no bearing on his reasoning without explaining why they were in the order. It also found errors of law and instructed the new judge to proceed consistently with the Supreme Court's decision in Moody and Fifth Circuit precedent. The reasons given for reassignment were that a judge who had merely excised the fabricated "facts" while keeping his result would have substantial difficulty setting aside his earlier views, and that sending the case back to him would diminish public esteem for the judiciary. Judge Haynes concurred only in the final paragraph — the vacatur, remand and reassignment.

What changed

Across the three cases the source of the fake case law moved. In 2023 it was a lawyer. In India in 2026 it was, on the bank's account, the tribunal's own research. In the United States in October 2026 it was a judge's law clerk using AI on a draft. The response escalated with it: a US$5,000 penalty on lawyers, then the setting aside of a ruling, then removing a judge from a case.

None of these courts banned AI. Judge Castel wrote that reliable AI assistance was not improper, and India's Supreme Court stated its resolve to use AI in aid of adjudication. What was at issue was the human responsibility for putting unchecked output into the record. The wider story is on the AI and society timeline; for a case over legal responsibility for a chatbot's own output, see the Character.AI lawsuit.