JudgeGPT? Artificial intelligence enters Philly judicial chambers, raising questions over accuracy.
A handful of judges in the U.S. District Court for the Eastern District of Pennsylvania have welcomed chatbots into their chambers. Penn researchers hope to keep the hallucinations at bay.

A group of federal judges in Philadelphia are using artificial intelligence to aid with cases being heard in the U.S. District Court for the Eastern District of Pennsylvania, raising new questions about the role of AI in the region’s justice system.
The judges, participants in a pilot program run by researchers at the University of Pennsylvania’s Carey Law School, have welcomed AI chatbots like ChatGPT into their chambers to schedule court hearings, draft jury instructions, and, in some cases, recall and summarize factual information about their cases. In all, around 25 judges in Pennsylvania and surrounding states are taking part.
The program, dubbed the PennLaw-Judiciary AI Testbed, provides ongoing training and support for judges and their staff as they learn about AI’s growing impact on the legal system.
Polk Wagner, a Penn legal professor who oversees the program, said participants are largely using the tools in their chambers to quicken administrative tasks, though some judges have used AI for more intensive work, such as summarizing elements of their cases and drafting documents.
“One of the challenges of being a judge is dealing with huge volumes of information and visualize it in a good way,” Wagner said. “These tools are really good at compiling information on a very large scale, if you know how to use them.”
Still, changing norms around AI behind the bench raise questions about what role the technology will play in courtrooms of the future. Generative artificial intelligence is known to hallucinate, producing answers to queries that sound correct but are factually inaccurate.
In 2025, a federal judge’s chambers in New Jersey was caught using the technology to draft an opinion that contained incorrect information and hallucinated quotes, stunning attorneys who spotted the errors. A Mississippi federal judge’s chambers made a similar blunder that year.
“People are aware of the dangers,” said Zachary Arbitman, a civil attorney and cochair of the Philadelphia Bar Association’s federal courts committee. “It’s another tool in the toolbox — you just have to use it appropriately.”
Arbitman said he and other attorneys continue to grapple with ethical questions over using artificial intelligence to aid their work. He was unaware that federal judges were exploring similar concerns through the Penn program.
The attorney said he is not aware of any court-wide rules on judicial AI use, noting that such technology policies typically come from judges themselves. In 2023, for example, U.S. District Judge Michael M. Baylson issued an order requiring that attorneys who submit legal filings written with the aid of AI must disclose that they have done so.
For U.S. District Judge Joshua D. Wolson, participating in the Penn program has allowed his chambers to experiment with a tool he predicts will become intertwined with courtroom work.
Wolson, who has served in the federal civil division since 2018, said he and his staff use artificial intelligence for a variety of tasks, such as drafting charging instructions for juries before deliberations, which are often lengthy and change slightly each trial.
“I may have the AI help me wordsmith while I have my law clerks work on the legal analysis piece of it,” Wolson said. “Frankly, it’s faster.”
Yet Wolson said he is also mindful of the technology’s risks, refraining from using AI for tasks like legal research, summarizing documents, or drafting opinions.
“Everyone — at least in this court — is aware of a concept called ‘human-in-the-loop,’” the judge said, referring to the idea that highlights the necessity of having a real person oversee artificial intelligence’s decision-making.
Those behind the Penn program, too, say they are focused on potential concerns as they continue to enlist judicial chambers to adopt the tools.
To that end, Wagner, the legal professor, said his team encourages judges and their staff to keep the scope of their AI usage as limited as possible.
For more intensive uses, he said, that means ensuring that their research queries focus on specific elements of a case — dates, locations, and keywords — rather than using chatbots as a search tool for legal knowledge or judicial precedent.
“These tools are pretty bad at legal research — even today,” Wagner said, noting that the wider set of information a chatbot is asked to crawl, the increased likelihood that it could produce an inaccurate answer.
The judges taking part in the program, which has grown from fewer than 10 chambers when it began in 2024, span a range of ages and prior experience with AI, according to Wagner. While a handful of clerks have become “skeptical and hesitant” over the technology’s potential to err, he said, judges have largely reported a positive impact.
“In our view, artificial intelligence is going to be integrated into legal practice from top to bottom,” he said. “The judges realize that, and they want to make sure they and their chambers are ready.”

























