Editor’s Note: This article was first published in Legaltech News.
When the misuse of generative AI in Mata v. Avianca Inc. made headlines in 2023, there was hope the widespread publicity about the AI “hallucinations” in the case might serve as a helpful warning.
Lawyers would learn the legal research lessons of Mata, understand how to use generative AI correctly, and that would be that.
Sadly, that has not been the case.
According to a database compiled by French researcher Damien Charlotin, as of late July, there were over 230 legal matters around the world where fictitious legal citations generated by generative AI became an issue.
However, not unlike a high-powered Lamborghini operated by a new motorist without a seatbelt or an owner’s manual, recent cases illustrate a common theme in this avalanche of AI legal research gone wrong: the technology isn’t usually the issue—it’s how it’s used.
Unprecedented No More
In the context of AI and machine learning, a hallucination refers to the generation of often plausible-sounding—but potentially inaccurate or fabricated—information.
When faced with the hallucinated legal citations filed with his court in Mata, U.S. District Judge Kevin Castel wrote, “This Court is presented with an unprecedented circumstance.”
He wasn’t wrong, but that was the bygone era of 2023.
Since Judge Castel’s May 2023 order to show cause in Mata, the issue is not only no longer unprecedented—it’s a continuing problem for courts around the world.
As Judge Castel wrote in his subsequent June 2023 sanctions order in Mata, “Technological advances are commonplace, and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.”
These existing rules, the legal research procedures they taught us in law school, and some generative AI training can go a long way in addressing the problem.
The July 17 sanctions order in ByoPlanet Int’l, LLC v. Johansson and related civil actions in the U.S. District Court for the Southern District of Florida highlights one of the more egregious examples of the misuse of generative AI.
From COVID to Court
ByoPlanet International has billed itself as “the leading manufacturer of electrostatic sprayers,” encouraging its customers to “embrace the future of disinfection” while experiencing “the power of advanced technology combined with superior craftsmanship to elevate your sanitization standards above the rest.”
With technology developed at the University of Georgia, ByoPlanet helped keep airlines, schools, hospitals, and other indoor spaces safe during the COVID-19 pandemic, according to the Georgia Department of Economic Development, and, according to the company, its cooling technology has been used by Major League Baseball’s Atlanta Braves, New York Mets, and New York Yankees, and several major universities.
However, the company’s experiences in court have not always been as successful.
In litigation in Canada with Promark Electronics, a court ordered substantial sanctions against ByoPlanet and its CEO, Richard O’Shea, and—as the legal battles with Promark continued south of the border in federal court in Florida—ByoPlanet was involved in a sanctions matter again.
This time, the sanctions were against ByoPlanet’s lawyer because of the way he misused generative AI and, the court ruled, misrepresented that use to the court.
Repeatedly Regurgitating Bad Law
Attorney James Martin Paul represented ByoPlanet and O’Shea in the Florida litigation, including state court cases and four matters in the U.S. District Court for the Southern District of Florida. He used hallucinated citations across eight different matters.
The way Paul misused generative AI—and a finding of bad faith—resulted in substantial sanctions, including the dismissal without prejudice of the four federal matters.
For an epigraph to his sanctions order, U.S. District Judge David Leibowitz used a quote from the late U.S. Supreme Court Associate Justice Antonin Scalia on this importance of the ethical obligations of candor and honesty in judicial proceedings.
Judge Leibowitz was making a point with the quote. It was an AI hallucination—generated by a July 7, 2025, ChatGPT prompt, “scalia quotes on candor.”
However, as usual, the sanctions in ByoPlanet were not sanctions for using generative AI to conduct legal research. Like Judge Castel in Mata, Judge Leibowitz was not saying lawyers should not use AI.
“While the use of AI by itself is not inherently suspect, wholesale reliance on AI without further inquiry or diligence by a lawyer is conduct which a court should deter, as lawyers must always conduct a reasonable inquiry,” Judge Leibowitz wrote.
The judge outlined Paul’s use of generative AI in both the state and federal litigation, noting that Paul had “repeatedly regurgitated” hallucinated cases across the matters, even after he was on notice that there were hallucination issues with his AI-generated citations.
Perhaps the most egregious violation?
The court noted that, in response to an order to show cause regarding the use of AI-fabricated case citations, Paul used—you guessed it—fabricated quotations.
Not helping matters, Attorney Paul informed the court that he and his paralegal used ChatGPT in drafting court papers, adding that his paralegal would draft the pleadings and briefs, which he would “tweak,” but not always review.
Rather than reducing any sanctions, Paul’s disclosure of the paralegal’s involvement threw fuel on the fire, with Judge Liebowitz noting that a paralegal drafting submissions to the court without attorney review constituted the unauthorized practice of law.
As a result of these transgressions, Judge Liebowitz dismissed all four matters without prejudice and without leave to amend, ordered Paul to pay opposing counsel’s attorney fees and costs, ordered Paul to attach a copy of the sanctions order to any case he files in the Southern District of Florida for the next two years, and the judge referred him to the Florida Bar for discipline.
No Bad Faith, But Still Bad
One of the cases Judge Liebowitz cited in ByoPlanet was another recent civil action in the Southern District of Florida, Versant Funding LLC v. Teras Breakbulk Ocean Navigation Enters., LLC, specifically, a May 20 sanctions order by his Southern District colleague, Chief U.S. Magistrate Judge William Matthewman, a respected authority on discovery law.
In Verdant Funding, Timothy Lord, a lawyer appearing pro hac vice, drafted a response that included a hallucinated case citation, and both Lord, an experienced attorney—but inexperienced generative AI user—and local counsel, Joel Bello, signed and filed the response. Not surprisingly, neither multiple opposing counsel nor Judge Matthewman were able to find the hallucinated case.
Approximately two weeks after opposing counsel’s reply informing them of the mysterious citation, Lord and Bello filed a notice withdrawing the citation, without stating the reason. On the same day, Judge Matthewman issued a paperless order requiring counsel to explain the non-existent citation.
Also, on the same day, Lord and Bello filed a response, informing the court that Lord called opposing counsel to apologize, that Lord was willing to personally reimburse opposing counsel for costs involved, that he realized there may be sanctions, and that Lord and Bello would institute procedures to help ensure such failure would not happen again.
As we’ve seen in the previous AI hallucination cases, Judge Matthewman did not blame the technology.
“In the Court’s view, there is nothing inherently wrong with an attorney properly and competently utilizing AI or any of its subsets to practice law or litigate cases,” Judge Matthewman wrote.
However, he added that the technology was evolving and was prone to provide hallucinated cases, laws, or facts. Thus, the judge wrote, “This is why close and careful attorney supervision, fact-checking, and citation-checking are absolute necessities when utilizing AI or any of its subsets.”
Although Judge Matthewman said Lord’s two-week delay in responding to opposing counsel’s notice of the hallucinated case was “an unnecessarily slow response to a serious misrepresentation,” he added, “the Court finds that Mr. Lord did not purposely attempt to mislead the Court. Nor did he intentionally submit a fake case citation.” In addition, the judge said local counsel Bellow was “certainly less culpable here,” adding that, not unlike Lord, Bello “did not engage in purposeful bad faith misconduct.”
Nevertheless, Judge Matthewman said the lawyers’ conduct was “careless, negligent, and reckless.” In addition to reimbursing opposing counsel’s fees and costs, the judge ordered them to take complete a CLE on artificial intelligence with an ethics component, and he fined Lord $1,000 and Bello $500.
Why ByoPlanet and Versant Funding Matter
As Judge Leibowitz noted in ByoPlanet, we lawyers of a certain vintage can remember “a bygone era when dinosaurs ruled the earth.” (That would be the 1990s.) During this “dark, pre-modern age,” a law student “often had to hold a volume of a legal reporter in one’s hands” and use citation services to ensure all cited cases were still good law. Even with the advent of electronic legal research tools, one still had to check citations.
Nothing has changed.
You still have to check citations.
As Judge Matthewman noted in Versant Funding, quoting U.S. District Judge Xavier Rodriguez (W.D. Tex.), a noted authority on the use of AI in the law, “Attorneys using AI tools without checking on the accuracy of their output are responsible for the consequences of incorporating inaccurate information into their work product.”
Back to that high-powered Lamborghini operated by a new motorist without a seatbelt or an owner’s manual, like generative AI, the motor car provides outstanding power, speed, and performance. There’s nothing wrong with the finely crafted motor car—depending on how you use it.
Likewise, Judges Castel, Leibowitz, Matthewman, and Rodriguez have all noted, there’s nothing wrong with lawyers using AI—depending on how they use it.
Legal guidance such as Federal Rule of Civil Procedure 11(b), comment 8 to ABA Model Rule of Professional Conduct 1.1 and its state counterparts provide the speed limits and seatbelts, and CLEs on the use of AI in the law provide the owner’s manuals. Judge Matthewman and Judge Rodriguez will provide one of those CLEs, joining Judge Allison Goddard, Justice Tanya R. Kennedy, and Dr. Victoria McCloud as they return for the 12th Annual Relativity Fest Judicial Panel on October 8.
Of course, another important takeaway when comparing ByoPlanet and Versant Funding is that lawyers’ conduct after a hallucination matters.
As Judge Matthewman wrote in Versant Funding, the lawyers “did not attempt to minimize their behavior. They did not attempt to cover up their error or obfuscate the issue. They accepted responsibility and apologized. Had they not done so, and had they attempted to cover up their conduct, the Court would be imposing much more serious sanctions in this case.”
When thinking of the roles of the lawyer and the technology, Judge Leibowitz summarized it succinctly in ByoPlanet, “Perhaps twenty years from now, AI will be flawless. Whenever that day comes, that flawless brief will only have meaning because the signature at the bottom does.”
The recipient of the 2024 Lifetime Achievement Award from the International Legal Technology Association (ILTA), David Horrigan is Relativity’s discovery counsel and legal education director. An attorney, award-winning journalist, law school guest lecturer, and former e-discovery industry analyst, David serves also as adjunct professor of law at the Duquesne University Thomas R. Kline School of Law. A former in-house counsel and reporter and assistant editor at The National Law Journal, David is the author and co-author of law review articles as well as the annual Data Discovery Legal Year in Review, David has been a contributor to Legaltech News for 23 years. His articles have appeared also in The American Lawyer, Corporate Counsel, The New York Law Journal, Texas Lawyer, The Washington Examiner, and others, and he has been cited by media, including American Public Media’s Marketplace, TechRepublic, and The Wall Street Journal. David serves on the Global Advisory Board of ACEDS, the Planning Committee of the University of Florida E-Discovery Conference, and the Resource Board of the National Association of Women Judges (NAWJ). David holds a juris doctor from the University of Florida Levin College of Law, and he is licensed to practice law in the District of Columbia. He is also an IAPP Certified Information Privacy Professional/US