AI hallucinations are no longer just lawyers’ problems as a US appeals court last week removed a federal judge over the AI curse of hallucinations.
The judicial hallucination occurred in Mississippi where an injunction issued by Judge Henry Wingate was vacated after the judicial procedure was labelled a “botch job”.
Unfortunately the Judge had misquoted state law and cited a case that does not exist . Never helpful, but all part of the traditional hallucination story where a lawyer asks a chatbot for authority, the chatbot obliges with invented cases, the lawyer files them, opposing counsel notices and red faces and appeals all round.
In the Wingate case, the Judge acknowledged that a law clerk had used AI to draft the decision.
The fix in this case did not fix it either as the replacement order still cited a non-existent 1974 decision. The appeals court found it at oral argument.
Wingate is not alone. In 2025, Judge Julien Xavier Neals of New Jersey withdrew an opinion in a securities case that contained fake quotes and misstated case outcomes. A person familiar with the matter told Reuters a temporary assistant had used AI.
The numbers keep climbing
Researcher Damien Charlotin’s public tracker listed 1,624 court decisions involving AI-fabricated content as of June 2026, up from roughly 200 a year earlier and a recent analysis puts lawyer AI misuse at over 2,000 matters worldwide.
Roughly half the tracked cases involve self-represented litigants, which is a reminder that judges now face fake law from both directions.
Other Jurisdictions
- England and Wales. In Ayinde v Haringey in 2025, the Divisional Court referred lawyers to their regulators and told managing partners and heads of chambers that the court would ask whether they had met their leadership responsibilities (Legal Cheek. The judiciary’s own AI guidance, updated in October 2025, stresses judges’ personal responsibility for everything issued in their name (Courts and Tribunals Judiciary).
- Australia. A King’s Counsel apologised to the Supreme Court of Victoria after filing fake citations and quotes in a teenager’s murder case. Justice James Elliott said AI use is unacceptable unless the output is “independently and thoroughly verified” (KIRO 7 / AP). Another Victorian solicitor lost his right to practise as a principal (Solicitor News).
- New Zealand. The courts issued separate generative AI guidelines for judges, lawyers and non-lawyers in December 2023 (NZ Lawyer). Disclosure of AI use is not required unless the court asks (Victorian Law Reform Commission).
Every common law system has now said the same thing: the human whose name is on the document owns every word in it. The Fifth Circuit has simply applied that rule to the bench, with consequences.
Who checks the judge?
That used to be a rhetorical question. Now it is a practical one, and the answer is you. When a lawyer files fake law, a judge can demand an explanation. When a judge issues it, as one law professor told Law360, lawyers cannot make the same demand (Law360). Your tools are the ordinary ones, used faster and more carefully.
If the ruling against you cites law that may not exist:
- Verify everything, not just the citations. Check that each case exists, that quotes appear in the source, and that holdings are stated correctly. Hallucinations increasingly look like real cases saying things they never said.
- Raise it promptly and politely. In the New Jersey case, a party’s letter flagging the errors led the judge to withdraw the opinion. A measured letter or application to correct the record often does more than an angry motion.
- Preserve it for appeal. If the error goes to the reasoning, put it on the record and in the grounds. The Fifth Circuit’s point was that hallucinated authority can infect the result, not just embarrass the author.
- Treat reassignment as the exception. The Mississippi case involved repeated errors and a botched correction. One bad citation in a judgment will not get you a new judge, and asking for one on thin grounds will not endear you to the old one.
- Mind your tone. Every lawyer reading this has a junior who has trusted a chatbot at 2am. Point out the problem, not the hypocrisy.
And the duty still cuts both ways. Courts are not softening on practitioners because judges have stumbled. Firms need a written AI policy, a cite-check of every authority including string cites, and a named partner accountable for both. The English court has already said it will ask.
The real lesson
The profession spent three years treating hallucinations as a competence failure by careless lawyers. The Fifth Circuit has made clear it is a systems failure that can reach anyone who signs a document without reading the sources. AI did not decide the Mississippi case, as plaintiffs’ counsel fairly pointed out (DeSoto County News). It just made it impossible for anyone to trust the ruling. In law, that is the same thing as getting it wrong.
