The seemingly limitless proliferation of cases in which lawyers have been caught letting fictitious AI-generated legal citations contaminate their briefs continues to amaze.

That’s not only because judges are fining more lawyers for their laziness, but because the publicity about these embarrassments has been inescapable.

Here’s one involving a dog named Kyra.

She’s a 16-year-old Labrador retriever who became the target of a nasty custody fight between a California couple after the dissolution of their domestic partnership. In the course of the lawsuit, one lawyer published two AI-fabricated citations in a filing. The opposing law firm didn’t catch the flaw and cited the same fake cases in its filings, including in a court order signed by a judge.

The case of Joan Pablo Torres Campos vs. Leslie Ann Munoz also points to how AI, touted worldwide as a labor-saving technology, has actually increased the workload in some trades and professions, like lawyering. For litigators, it has created a new imperative: ferreting out citations that have been fabricated by AI bots in their own court filings — and their adversaries’.

I’ve written before about the proliferation of AI-generated fabrications infiltrating legal filings and even legal rulings, despite the advice drilled into the heads of even law students about making sure that their citations to precedential cases are accurate. But the wave keeps building: A database of AI hallucinations maintained by the French researcher Damien Charlotin now numbers 1,174 cases, of which some 750 are from U.S. courts.

That’s almost certainly a conservative count. Most AI fabrications may not even come to the attention of litigants or judges, especially in state courts.

“For every case that talks about this, my guess is that there are many that aren’t visible,” says Eugene Volokh of UCLA law school and the Hoover Institution, who keeps a weather eye on AI-related courthouse developments. He believes there may be thousands escaping notice.

AI has introduced mistakes that were never seen in the past. “Most lawyers grew up in a time when you could expect the other side to spin and even to lie about the record some of the time, but just lying or making a mistake about the existence of a case was basically unheard of up until a few years ago,” Volokh told me. “That’s because there would be no source of hallucinations — maybe you’d get the citations slightly wrong or you mischaracterized or misquoted them, but to talk about a case that doesn’t exist — that didn’t happen. Now it happens a lot.”

The judiciary is getting increasingly nervous about AI fabrications becoming part of the judicial record. “Reliance on fake cases…seriously undermines the integrity of the outcome and erodes public confidence in our judicial system,” an appelate judge stated.

Therefore, he added, “it is imperative for both the court and the parties to verify that the citations in all orders are genuine….This is especially vital with the increasing incidence of hallucinated case citations generated by AI tools.”

Judges are still reluctant to bring down the hammer for AI-fabrications if lawyers acknowledge their fault and “throw themselves on the mercy of the court,” Volokh says. But they’re getting tougher on lawyers who deny their reliance on AI or try to shift blame.

As recently as Monday, federal Magistrate Mark D. Clarke of Medford, Ore., ordered the attorneys representing the plaintiff in a civil lawsuit to pay more than $90,000 in legal fees, on top of an earlier sanction of $15,500 imposed on one of the lawyers, for incorporating 15 fabricated case citations and eight misquotations into case filings.

Clarke also dismissed the $29-million lawsuit, which arose from a ferocious dispute among the sibling heirs to an Oregon winery fortune, with prejudice, so it can’t be refiled. It was an extraordinary punishment, Clarke acknowledged — and the largest penalty imposed in any case in Charlotin’s database.

“In the quickly expanding universe of cases involving sanctions for the misuse of artificial intelligence, this case is a notorious outlier in both degree and volume,” Clarke wrote. Among other faults, he noted, the plaintiff’s lawyers never adequately fessed up to their wrongdoing. “If there was ever an ‘appropriate case’ to grant terminating sanctions for the misuse of artificial intelligence,” he wrote, “this is it.”

That brings us back to the custody battle over Kyra. The case originated in 2024, two years after a family court judge in San Diego dissolved the domestic partnership of Joan Torres Campos and Munoz. The dissolution order allowed them to keep their own property, but didn’t mention the dog, who lived with Munoz.

Torres Campos subsequently sought shared custody of Kyra and visitation rights. (Pet custody battles have long been a cultural fixture: Film aficionados might recognize this case’s similarity to the custody fight over the wire-haired terrier Mr. Smith in the 1937 Cary Grant/Irene Dunne vehicle “The Awful Truth,” surely the funniest movie ever made by Hollywood.)

Munoz rejected Torres Campos’ request, arguing that he didn’t really care about the dog, but only aimed to harass her. A family court judge sided with her, but Torres Campos appealed.

In her initial reply to Torres Campos, Munoz’s lawyer, Roxanne Chung Bonar, cited California cases from 1984 and 1995 that she said supported her client’s refusal to grant visitation rights.

Both case citations were fictitious. The 1984 case, Marriage of Twigg, didn’t exist at all; Bonar’s citation pointed to a criminal case that had “nothing to do with pets or custody determinations,” California Appellate Judge Martin N. Buchanan wrote for a unanimous three-judge panel, upholding the family court judge . The second reference was to Marriage of Teegarden, which was handed down in 1986, not 1995, and also had nothing to do with the issue at hand.

Things only got more complicated from there. Torres Campos’ lawyer, in a reply brief and a subsequent proposed court order, didn’t mention that Twigg and Teegarden were fabricated cases, perhaps because the lawyer hadn’t checked the references personally. The family court judge signed the proposed order, including the fake citations, resulting on their infiltration into the official record. (Although Torres Campos’ lawyer drafted the proposed order, it actually rejected his lawsuit.)

It was only in the course of appealing the family court ruling did Torres Campos’ lawyer mention that the two cited precedents were “invented case law.”

There was one more turn of the screw: In responding to Torres Campos’ appellate filing, Bonar “doubled down,” Buchanan wrote. Bonar insisted that Twigg was a “valid, published precedent” and added three more purported citations to the case. All were “just as phony as the original citation,” Buchanan noted.

Bonar even taunted Torres Campos’ lawyer for his “failure to conduct basic legal research” to verify the ostensibly genuine precedents, adding that his “inability to locate them underscores the incompetence that led to his appeal’s dismissal.”

Where did these references come from? It turned out that the Twigg reference originally came from a Reddit article written by an Oregon blogger and animal rescuer who posts under the name “Sassafras Patterdale,” in which she cited the fictitious case in a post about pet custody battles. Munoz had received the article from a friend and passed it on to Bonar. Both of them assumed that everything in it was accurate.

According to the appellate ruling, the additional citations to Twigg don’t appear in the Reddit post. Bonar never explained where they came from. She did concede, however, that the fictitious citations “‘may have’ come from her use of AI tools,” Buchanan noted. He sanctioned her with a $5,000 fine, largely because she did not initially acknowledge that her citations were fake and tried to shift blame to her opposing counsel.

Although the appeals judges could have awarded the case to Torres Campos due to Bonar’s performance, they declined to do so — because Torres Campos’ lawyers hadn’t checked their opposing counsel’s citations themselves. At this stage, Munoz still has custody of the dog and the lawsuit is essentially over, according to Torres Campos’ attorney, David C. Beavens of San Diego.

Beavens says he took the case because he hoped to use it to obtain judicial clarification of a state law enacted in 2019, which authorized courts to issue orders regarding the ownership and care of pets in divorce cases. The appellate judges, sidetracked by the AI issue, never touched on that. But Beavens says he agreed with the panel’s position AI fabrications have become such a problem in court that “we need to hold everyone accountable” — lawyers on both sides of a case and the judges as well.

Bonar told me that she was not challenging the sanction but declined to comment on it further.

I did ask Bonar if she had any advice for other lawyers tempted to use AI in their work. “Yes,” she said: “Verify all third-party sources.”