An Oklahoma district judge admitted to state investigators that he cited at least two nonexistent legal cases in a child custody order, relying on ChatGPT to generate them. The admission, detailed in a letter from the local district attorney, exposes how generative AI tools can fabricate legal precedent and underscores the professional risks for lawyers and judges who use the technology without verification.
Stephens County District Judge Lawrence Wheeler denied a request to compel a psychological evaluation in a custody case last November. His written order referenced multiple prior Oklahoma cases that he claimed supported his decision. Those cases were never decided by any court. The Oklahoma Council on Judicial Complaints referred the matter to the Oklahoma State Bureau of Investigation (OSBI), which interviewed Wheeler.
According to an August letter from Stephens County District Attorney Jason Hicks to the state Attorney General's office, Wheeler told OSBI agents that ChatGPT supplied at least two cases that "do not exist." Wheeler said he wrote the order himself and used the AI tool only for research.
No criminal charges, but disciplinary questions remain
The Attorney General's office reviewed the OSBI findings and concluded that "the evidence does not support a criminal prosecution." A spokesperson noted that the Oklahoma Supreme Court holds the ultimate authority to discipline judges. The attorney representing Judge Wheeler had no comment when contacted by Nexstar's KFOR.
Tim Gilpin, an attorney and former Assistant State Attorney General, said the incident reflects a broader problem. "Even here in Oklahoma, judges are very upset when attorneys use AI in their briefing, but don't clearly document in the brief," Gilpin said. He added that his own experience with AI tools reveals a recurring flaw: "AI, when I use it, it almost always has a false case or cites a case that doesn't stand for what AI says it does."
How the legal profession is confronting AI hallucinations
Large language models do not search a database of real cases. They predict text patterns based on training data, which means they can generate citations that sound plausible but are entirely invented. This phenomenon, often called hallucination, has already ensnared practicing attorneys in other states who filed briefs containing fake case law. A judge using such output in a ruling raises the stakes further.
Gilpin said the episode surprised him because it showed blind trust in the technology. "It's programmed to try to give you the answer that you're looking for and in all likelihood, the answer that you want," he said. "So there's the first warning is that you need to take a good look at what it's giving you with a big grain of salt."
For legal professionals seeking to understand these tools without falling into similar traps, structured training can help distinguish productive use from reckless reliance. Resources such as AI for Legal Professionals Courses cover the mechanics of generative AI and its limitations in document review and research contexts. An AI Learning Path for Paralegals offers guidance on integrating AI into legal workflows while maintaining verification standards.
Why this matters for legal professionals
This incident signals that courts are now scrutinizing AI-generated content at every level of the system. A judge's admission that fabricated cases appeared in a ruling erodes trust in the bench and creates appeal risks for affected parties. For attorneys, the takeaway is practical: any AI tool used for research must be treated as a starting point, never as authority. Gilpin framed the boundary clearly. "The trick is, AI is a tool for us lawyers. It's not the answer. It's a tool to arrive at the answer," he said. "If we use it as the beginning and the end of the question, that's when people come into trouble."
Your membership also unlocks:
