AI Hallucinations Land Cleveland Attorney In Hot Water

AI in the Courtroom A Cautionary Tale

The promise of artificial intelligence is to streamline work and boost efficiency, but for one Cleveland defense attorney, it has become a source of professional peril. William Norman is facing potential sanctions in two Ohio counties after court filings submitted by his office were found to contain fabricated information generated by ChatGPT. The incidents serve as a stark warning about the risks of using generative AI in the high-stakes legal field without meticulous human oversight.

William Norman faces potential sanctions in two counties after using ChatGPT in court documents, which produced fabricated information and left a prompt visible in one filing. (AP Photo/Kiichiro Sato)

The documents, filed in Cuyahoga County Common Pleas Court and the 11th Ohio District Court of Appeals, have drawn sharp criticism from judges and prosecutors, raising serious questions about the ethical use of AI in legal practice.

The Case of the Hallucinated Quotes

The first issue arose in an appeal for Malikhi Coleman, a man convicted of murder in Ashtabula County. Norman admitted in a court record that a paralegal in his practice used ChatGPT to help draft a motion. The resulting document was deeply flawed, containing quotes that were completely fabricated and improperly attributed to the prosecutor during the original trial's closing statements.

The appellate court noted the "gravity of the violations" and demanded an explanation. In response, Norman's attorney, Joseph Dunson, acknowledged the error, stating, “ChatGPT hallucinated inaccurate trial transcript quotes.” He insisted the mistake was not made in bad faith and that Norman had taken accountability for the oversight.

Facing the Consequences

In an attempt to make amends, Norman implemented new office protocols to prevent future AI misuse and voluntarily paid $2,000 to the Ashtabula County prosecutor’s office. However, prosecutors argue that this was not a simple mistake. They contend that Norman's conduct was a willful violation of professional rules, which mandate that all legal filings be grounded in fact. They have pointed to national cases where courts have imposed sanctions, from public reprimands to disqualification, for similar AI-driven misconduct.

“Mr. Norman has committed significant misconduct in this case, and there are unfortunate indications that it may happen again,” wrote Assistant County Prosecutor Dane Hixon. The appellate court has yet to rule on potential sanctions.

History Repeats Itself in Cuyahoga County

While the first case was still under review, a new problem emerged. Prosecutors in Cuyahoga County moved to sanction Norman for a separate filing. This document, aimed at withdrawing a client's guilty plea, was littered with AI-generated errors. It included citations to legal cases that do not exist and, incredibly, contained a leftover AI prompt: “Would you like me to draft the next argument section...?”

This second incident suggested to prosecutors that the first was not an isolated lapse in judgment. They argued that Norman had failed to learn from the prior warnings and was continuing to submit sloppy, unverified work.

An Unbelievable Explanation

Norman's attorney offered an explanation for the Cuyahoga County filing, claiming that the wrong file was mistakenly uploaded and that steps were taken to correct it. This explanation was met with skepticism.

“For this to be happening while there is a pending hearing on sanctions is nearly beyond belief,” Hixon, the Ashtabula prosecutor, wrote in response. A hearing on the matter is scheduled for December 10 before Common Pleas Judge Brian Mooney. These back-to-back incidents underscore the critical need for legal professionals to verify every piece of information, especially when it comes from an AI that is known to "hallucinate."