Heidi Hafer was a stranger to appellate law when she drafted a nearly 50-page statement last May to Texas’ Fifth District Court of Appeals. Hafer — who works as a corporate attorney — was trying to challenge a ruling to take a million dollars’ worth of jewelry she says were gifts to settle her family’s debt.
To support her case, Hafer cited 31 court decisions, including: Macy’s Texas, Inc. vs. D.A. Adams & Co., a 1979 case out of a Tyler court; Stephens vs. Beard from 1958; Gaston vs. Monroe dated 1896; and a 1996 San Antonio case titled Estate of Malpass vs. Malpass.
But no one — not the appeals court in Dallas, the creditor’s lawyers nor Hafer herself — could find those four cases.
That’s because they never existed.
Related:Trump draws criticism with AI image of himself as the pope ahead of papal conclave
Now, Hafer is facing possible sanctions for allegedly using generative artificial intelligence to prepare the court filing. In a four-page report to the appeals court last month, Hafer took full responsibility for the errors.
“This is my first brief and I really didn’t know how to address it all, so I did the best I could,” Hafer testified at a hearing Thursday morning when asked why she didn’t immediately address accusations that the cases were fake.
Hafer has been a licensed practitioner in Texas since 1999 and has not been disciplined by the state bar. She is listed as chief general counsel on a website for a Dallas-based company that deals with artificial intelligence.
John Browning — a former appellate judge and chair of the State Bar of Texas’ Taskforce for Responsible AI in the Law who represented Hafer at the hearing — said Hafer has not tried to obfuscate her mistake nor “tried to blame the cat for eating her homework.”
Addressing a panel of three justices, Browning said his client has taken full responsibility. “She admits that this is an embarrassing chapter in her professional career. She has expressed the contrition that I would expect, and she has taken remedial steps to better herself as an attorney and to avoid this in the future.”
It’s unclear whether Hafer used generative AI unwittingly: She told the justices she used Google to search for common law on gifts and did not recall using any other AI-powered tool. Google has its own AI chatbot and sometimes populates AI-generated summaries of queries at the time of search results.
Related:Fight over ‘truth’ in AI political ads heats up as Texas House passes transparency bill
Outside of the courtroom, Browning told The Dallas Morning News, “artificial intelligence is all around us. … A lot of people don’t realize how pervasive AI is in our lives.”
Hafer and lawyers for the creditor, JGB Collateral, LLC, declined to comment.
On the heels of the American Bar Association’s guidance on AI, Texas’ state bar issued its own framework, cautioning lawyers to verify the accuracy of AI-generated content. At least three federal judges in North Texas require attorneys to verify that language drafted by AI will be checked for accuracy, according to an online database of judicial orders. Similar policies have been enacted in state district courts from Bexar to Wichita counties, the database shows.
“Lawyers have a duty of competence, and that includes technological competence,” said Korin Munsterman, a professor at the University of North Texas at Dallas College of Law who researches the intersection of law and technology.
Munsterman added: “It is imperative that lawyers rely on their expertise, and not blindly accept a GenAI’s output just because it came out of a computer.”
Related:AI company relocating U.S. headquarters from San Francisco to North Texas
AI offers attorneys a way to streamline complicated or tedious legal tasks and perhaps cut down on billable hours, and 68% of professionals at law firms reported using generative AI at least once a week, according to a 2024 survey. The major drawback, however, is AI’s propensity to spit out wrong or misleading information. Researchers at Stanford University found that even AI tools designed to help lawyers spawned incorrect information between 17% and 34% of the time.
This isn’t the first time the technology has led lawyers astray: An attorney was accused of using AI to challenge a man’s bail in a Brazos County case. Lawyers or firms in New York and Colorado have also been reprimanded for using fake judicial decisions or cases conjured by services like ChatGPT. Michael Cohen, former fixer for President Donald Trump, mistakenly gave his lawyer AI-hallucinated cases but avoided sanctions.
Related:Why Big Tech’s sudden obsession with DeepSeek’s new AI is a big deal
Browning advocated Hafer not be further punished for what he called an “honest mistake.”
“Hafer was not attempting to mislead the court,” Browning explained. “Valid case law does exist to support the premises she was arguing — she simply cited nonexistent (cases),” he said.
It’s unknown when the justices may issue their decision.
