Judges in one of the nation’s largest court systems have started using artificial intelligence, testing a tool that can rapidly distill hundreds of pages of legal motions and use samples of a jurist’s writing style to help reach conclusions and even draft tentative rulings.
The program, which launched last month, gave half a dozen Los Angeles County civil court judges access to AI software called Learned Hand. Although it could prove critical in a shorthanded court system that is facing a workload crisis on many fronts, the announcement has also drawn concern from some members of the county’s legal community who fear the technology could create errors and erode public trust in the legal system.
Court officials say judges in the pilot program are “required to review and edit the draft before adopting tentative rulings” generated by Learned Hand, and they touted the new effort to use technology to assist with basic judicial tasks and clear case backlogs.
“Judicial officers have long been supported by research attorneys and law clerks who assist with summarization, legal research, analysis and drafting assistance,” said Rob Oftring Jr., the court’s chief spokesman. “This assistance does not supplant the judicial officer’s independent role in decision-making.”
Shlomo Klapper, chief executive and founder of the company behind Learned Hand, said it is already being used by court systems in 10 states. The Michigan Supreme Court began using the software last summer to review applications for permission to appeal in civil and criminal cases, according to a spokesman for the court.
Klapper described the AI tool as a co-intelligence, akin to a “judicial sous chef,” that will support members of the bench without supplanting them.
Klapper, who worked as an attorney and federal law clerk before starting Learned Hand in 2024, said it’s a necessary aid for a judiciary drowning in a “paper blizzard,” especially with public access to AI models such as ChatGPT leading to more self-represented litigants filing cases in civil court.
“This is what’s giving me such urgency. We need to build the right tools so courts are equipped to deal with this tsunami,” he said. “The system is drowning and the flood hasn’t even started.”
Los Angeles County Dist. Atty. Nathan Hochman expressed some concern with the county’s plan. He said AI could be useful in cutting down the time judges spend on repetitive tasks such as assessing motions for summary judgment in civil court, which often cite the same case law and paragraphs over and over again. But he described the use of AI to generate rulings as “problematic.”
“Even when a judicial assistant or a law clerk comes up with a tentative on which position the judge should take, before the judge has taken their own position, that greatly influences what the judge’s position should be,” Hochman said, warning the AI-generated tentative ruling could predispose a judge before they conduct a legal analysis.
Acknowledging growing public anxiety about the integration of AI into different facets of society, Klapper turned to pop culture to assuage fears. He said he’s not building Skynet — the artificial intelligence that brings about the end of days in the “Terminator” films — but something similar to Jarvis, Iron Man’s affable computer assistant.
“I don’t come from a disruptive mindset. … I’m here to build,” he said.
AI has caused incidents in the legal system that critics say justify concern. Last year, a Los Angeles attorney was fined for submitting a filing full of legal citations that were hallucinated by ChatGPT. Last month, a federal prosecutor in North Carolina resigned after submitting a filing that was almost wholly produced by the same artificial intelligence.
But a Reuters survey conducted last summer also found more than 70% of respondents believe AI is a force for good in the legal field that can drastically reduce the amount of human work hours put into tedious tasks, including reviewing lengthy documents.
Klapper says Learned Hand has extensive guardrails to prevent the AI from inventing precedents and making other major mistakes. He said the program uses a fact-checking process called “Deep Verify,” which interrogates every sentence of a generated order to ensure the facts laid out match up with case law citations, which are available for review via hyperlink.
“We don’t just tell the judges to trust us,” he said. “We say you can actually verify it yourself and see from particular sources where things are coming from.”
One L.A. County judge, who spoke on the condition of anonymity because California court regulations largely bar judges from speaking with the media, echoed Hochman’s concern that an AI-generated tentative ruling could create bias.
“Even if you don’t necessarily adopt the AI’s tentative decision, psychologically that has become your reference point and any decision-making engaged in thereafter could be predicated on it,” said the judge, who is not part of the pilot program and has not used Learned Hand.
Judges would not have to disclose whether they used the program to aid in research or in the generation of a ruling, according to court officials. David Slayton, the L.A. County Superior Court’s chief executive, said that state court rules require judges to consider disclosing the use of generative AI in their process, but that there is currently no rule that would force them to do so.
The county’s contract with Learned Hand will see the pilot program stretch into early 2027 at a cost of a little over $300,000. The pilot program will see the tool largely used to review and summarize a wide array of civil court motions — including motions for summary judgment and motions for approval of class-action settlements — though it could have limited applications in the future in criminal courts for applications for postconviction relief, according to the contract. The software is not being used in the criminal courts.
Klapper said he understands why there might be some hesitance among judges or the public, but recalled cases sitting on his desk for nearly a year because he didn’t have five spare hours to read through voluminous motions. Learned Hand, he said, is not meant to replace judges but rather give them more time to actually make decisions rather than sit buried under impossible caseloads.
“There is no reason on fear that any technology company on earth, much less my own, should be making consequential decisions for the public,” he said.
