A Georgia lawyer who relied on ChatGPT while drafting court filings was ordered to personally pay more than $25,000 after a judge found that several of the cases cited in his briefs did not exist, turning what should have been an ordinary civil dispute into a prolonged fight over fabricated legal authority.
The hearing in Clayton County State Court, covered by YouTube channel The Court Docket Network, centered on attorney Tristan Gillespie, whose filings on behalf of plaintiff Casten Cruz contained multiple citations that Judge Tammi Long Hayward said were not actual Georgia cases and therefore could not be relied upon as legal authority.
Defense attorney Luke Kennedy had identified the false citations and later asked the court to sanction Gillespie for the time and expense caused by responding to them.
Gillespie acknowledged that he had relied on an artificial intelligence tool and said the experience had taught him the risks of using large language models for legal research, but Hayward rejected the idea that waiting for the Georgia State Bar to decide what to do should delay consequences in her courtroom.
Gillespie Says He Trusted The AI
At the hearing, Gillespie told Hayward that the technology was relatively new to him and said he now understood that it was not enough to ask a large language model to verify case citations.
He explained that, after the problem surfaced, he obtained proper legal research software and now relies on that instead.
Gillespie argued that the court should wait to see how the state bar responded before deciding whether sanctions were appropriate, saying the bar might consider the explanation he had offered for how the citations ended up in the filings.
Hayward was not persuaded.
She told him that whatever disciplinary decision the bar eventually made was separate from the damage already done to the case before her.
“You filed a pleading with the court that was replete with cases that did not exist,” Hayward said, adding that lawyers are responsible for the truth and reliability of anything they submit.
Judge Says The Brief Was “Essentially Fiction”
Hayward said Gillespie’s filings had forced the court to stop dealing with the actual dispute between the parties and instead spend time investigating legal authorities that turned out to be false.
She described one of the briefs as “essentially fiction.”
The judge said the filings had delayed the litigation, consumed court resources and undermined her confidence in Gillespie’s ability to continue representing the plaintiff by himself.
She also said she was relieved that attorney Taylor Middleton had entered the case on Cruz’s behalf because, after what had happened, Hayward was no longer comfortable proceeding with Gillespie as the sole lawyer for the plaintiff.
That distinction is important because the judge did not treat the problem as a harmless technical mistake. In her view, fabricated citations disrupted the entire process because every legal argument in a brief depends on the court being able to trust that the cases cited actually exist and say what the lawyer claims they say.
Kennedy Says The Problem Went Beyond One Filing
Defense attorney Luke Kennedy told the court that the issue was not limited to a single bad brief.
He said several filings contained the same type of problems and argued that his clients had spent significant time and money responding to documents that relied on fictitious or misleading authorities.
Kennedy asked the court not only to award attorney fees but also to dismiss the case, arguing that the conduct had prejudiced his clients and pushed the litigation close to a year without meaningful progress on the underlying claims.
He also accused Gillespie of failing to be candid when the problem was first raised.
According to Kennedy, Gillespie initially responded as though he did not understand how the false citations had appeared, even though he later acknowledged that the material came from AI-generated research.
Kennedy argued that bringing in a new attorney should not simply erase what had already happened.
Middleton opposed dismissal, saying Cruz should not lose her case because of misconduct she did not know about and had no role in creating.
Hayward agreed with that part of the argument.
She said it would be unfair to punish the plaintiff for filings prepared by her attorney, particularly if Cruz had no idea that false legal authorities were being submitted in her name.
Judge Draws A Line Between AI Assistance And Responsibility
Hayward then addressed the broader issue of artificial intelligence directly.
She said judges had recently received training about AI and that the central lesson was simple: the person using the tool is still responsible for the result.
Hayward acknowledged that AI can help with workload, scheduling and document preparation, but she said an open-source tool such as ChatGPT is not a substitute for legal research software designed to verify cases and statutes.
She compared it to a basic computing rule she remembered from earlier technology eras: “garbage in, garbage out.”
Her point was not that lawyers can never use AI.
It was that they cannot hand responsibility for accuracy to the software.
That is likely the most important lesson from the hearing because the technology itself was not sanctioned; the lawyer was. The court treated AI as another tool, which means the professional using it remains responsible for checking the work before filing it.
The Sanction Comes To $25,498
When Hayward turned to the amount of the penalty, Kennedy said his firm had incurred $25,498 in fees and costs up to the point when the sanctions motion was filed.
The judge ordered Gillespie to pay that amount personally.
She made clear that neither Cruz nor Gillespie’s former law firm should bear the cost.
The sanction, Hayward said, belonged to Gillespie because he was the one who chose to rely on the AI-generated material without independently checking whether the cited cases were real.
Gillespie objected that the amount seemed excessive because not every hour billed by Kennedy could necessarily be tied directly to the AI issue.
Kennedy said he could further itemize the fees if the judge wanted.
Hayward declined.
She said she might have viewed the matter differently if there had been only one isolated filing, but the pattern across several briefs convinced her that the full amount was justified.
The judge ordered the $25,498 to be paid within 120 days.
The State Bar Is Also Reviewing The Matter
Hayward told the attorneys that she had already forwarded the issue to the Georgia State Bar.
At the time of the hearing, she said the bar had confirmed receipt and the matter was being processed by its general counsel.
That review is separate from the court’s sanction.
Hayward emphasized that she was not going to wait for the bar because her responsibility was to address what had happened in her own courtroom, including the wasted time, delayed proceedings and unreliable filings.
Middleton, who had recently taken over the plaintiff’s representation, said she intended to get the case back on track and continue addressing discovery and other pending issues.
The hearing then shifted into the underlying civil dispute, where Kennedy raised additional concerns about missed discovery responses, deemed admissions and an affidavit he said contained conflicting facts.
Hayward again expressed hesitation about relying on anything previously filed by Gillespie, saying the court no longer had confidence in the validity or legal basis of his work product.
The Case Shows Why Verification Still Matters
The sanction stands out because the problem was not simply that AI made a mistake.
The problem was that the mistake reached the court in the form of formal legal filings without being caught first.
Gillespie told the judge he had learned from the experience and had changed how he conducts legal research, but Hayward’s response made clear that education after the fact did not erase the burden already placed on the court and the opposing party.
Kennedy’s argument was similarly practical: his clients had spent real money responding to briefs built on authorities that either did not exist or did not support the propositions for which they were cited.
For lawyers, that makes this case less about whether AI belongs in legal work and more about where responsibility ends.
Hayward’s answer was direct.
A lawyer can use new technology, but the lawyer’s name remains on the filing, and when fictional cases make their way into court, the consequences belong to the person who submitted them.
Ed spent his childhood in the backwoods of Maine, where harsh winters taught him the value of survival skills. With a background in bushcraft and off-grid living, Ed has honed his expertise in fire-making, hunting, and wild foraging. He writes from personal experience, sharing practical tips and hands-on techniques to thrive in any outdoor environment. Whether it’s primitive camping or full-scale survival, Ed’s advice is grounded in real-life challenges.
