Two of Southern California’s largest court systems, including Riverside County Superior Court, are testing an artificial intelligence program designed to help judges and court attorneys research cases, summarize motions and draft legal documents — with possible future use in criminal matters where liberty is at stake.
The tool, created by a company called Learned Hand, is being piloted in Los Angeles and Riverside counties. It functions as a kind of AI law clerk, drawing on language models from Anthropic, OpenAI and Google to produce research memos, draft orders and assist with tentative rulings.
So far, court officials say the technology is being used mainly in civil and probate matters. But contracts and emails obtained by CalMatters show the courts have discussed or authorized potential testing in criminal, family and probate divisions, raising concerns among judges, public defenders and prosecutors about accuracy, bias and transparency.
In Riverside County, the Superior Court has a $10,000 agreement with Learned Hand to test the program. Seven civil and probate attorneys have access to it, according to court officials, primarily to see whether it can help staff with research and other time-consuming tasks. Riverside court officials said the tool is not being used to draft tentative rulings.
Los Angeles County Superior Court has a much larger contract, worth about $314,000, that includes a plan to evaluate whether the tool could be used beyond civil cases, including in criminal, family and probate court. Los Angeles court officials told CalMatters the program is still in an early testing phase and remains months, if not years, away from broader implementation.
The use of artificial intelligence in courtrooms has become increasingly controversial as generative AI systems have been shown to create false citations, misstate facts and reflect biases found in the data used to train them. Those risks are especially sensitive in criminal cases, where court decisions can affect whether evidence is admitted at trial, whether a conviction stands, or whether a person remains incarcerated.
Los Angeles County District Attorney Nathan Hochman said he is especially concerned that the Los Angeles contract allows the technology to be used for motions to suppress evidence and motions for post-conviction relief — two significant areas of criminal law.
“When you’re dealing with someone’s liberty — as opposed to in the civil setting, which is everything other than liberty — the stakes couldn’t be higher,” Hochman told CalMatters. “I don’t want to take the chance, particularly in a criminal case, that AI happens to get it wrong.”
He said an error could cut both ways: violating a defendant’s constitutional rights or undermining a lawful prosecution.
Court officials in both counties have adopted generative AI policies, as required by the state Judicial Council before courts use the technology. But those policies require disclosure only when a motion, decision or other document is written entirely by generative AI. That means litigants may not be told if AI was used in part to research or draft material connected to their case.
Both Los Angeles and Riverside courts declined to say whether parties are being informed that the tool may be tested on their cases. A Los Angeles Superior Court spokesperson said testing is being done on motions that have already been decided and outside live case environments, though the contract permits testing on active matters.
“It is important to note that even with successful evaluation and thorough testing, the Court remains several months, if not years, away from implementing this type of tool,” the spokesperson said.
In Los Angeles County, six superior court judges and their research attorneys are using Learned Hand primarily to conduct legal research, summarize motions and assist with tentative rulings in civil matters, according to David Slayton, the court’s executive officer. Slayton said the court will not expand the tool into other divisions until court leaders are satisfied it can be used appropriately.
“The court is being very deliberate and careful about how we use technology like this,” Slayton said.
A court spokesperson later said the program is being evaluated under the same standards applied to human law clerks and research attorneys, including accuracy, legal analysis, neutrality and reliability. The court did not provide detailed criteria for how it will decide whether the technology is safe enough for use in criminal or family law matters.
Some judges have expressed alarm about how far the technology could go. One Los Angeles County Superior Court judge, who spoke anonymously because of judicial conduct rules, said colleagues were told during a recent luncheon that the tool might one day assist with petitions filed under California’s Racial Justice Act.
The 2020 law allows people to challenge criminal convictions or sentences if they believe racial bias affected the outcome. Courts across California are now processing a growing number of such claims, many filed by people in state prison.
According to the judge, court leaders discussed the possibility that AI could help generate tentative decisions for judges reviewing whether those petitions should be denied or move forward. The judge warned that such use could shape judicial thinking too early in the process.
“It’s an extremely perilous road to go down,” the judge said. “Putting aside the inaccuracy, which will be a significant concern, it dehumanizes the whole process.”
A second Los Angeles judge also told CalMatters they would not trust the tool to summarize or analyze a Racial Justice Act petition.
Los Angeles court officials and Learned Hand founder and CEO Shlomo Klapper denied that there are current plans to use the program for Racial Justice Act petitions. A court spokesperson later confirmed the contract would allow such use, but said that possibility “has not commenced in any way.”
Klapper said any Racial Justice Act application would need to be developed with the court and tested for bias. He acknowledged that criminal cases require heightened caution.
“It’s a very fraught decision,” Klapper said. “Extremely high stakes — a scenario where I understand people might be very concerned.”
Public defenders said those concerns are well-founded. Elizabeth Lashley-Haynes, a deputy public defender in Los Angeles County who specializes in Racial Justice Act litigation, said using AI to review such petitions would be “highly problematic and bordering on unethical.”
She said Racial Justice Act claims often involve subtle context, historical patterns and language with racial meanings that are difficult even for trained legal professionals to evaluate.
“They’re like nothing else in the legal system that has ever really been done,” Lashley-Haynes said.
Concerns about bias are not hypothetical. AI systems can reproduce or amplify patterns found in their training data, including racial and gender bias. Studies and investigations have found bias in large language models, predictive policing systems and criminal risk assessment tools.
Generative AI has also already caused problems in the legal system. Researcher Damien Charlotin has documented hundreds of instances in which litigants, attorneys or judges relied on false or faulty AI-generated information, including nearly 90 cases in California state or federal courts since August 2024. Last year, a Los Angeles-based attorney was fined $10,000 for citing cases that did not exist. More recently, AI-related errors were reported in several cases handled by prosecutors in Nevada County.
Most known incidents have involved lawyers or self-represented litigants, but legal scholars have warned that judges are likely to make similar mistakes as AI becomes more common. In recent months, the U.S. Senate examined apparent AI use by federal judges in Mississippi and New Jersey after rulings contained serious factual errors.
Klapper, a former federal appellate clerk who previously worked for the surveillance technology company Palantir, argues that courts must consider AI because they face growing backlogs and limited staffing. He said the goal is not to replace judges but to give them better support.
“I think the only solution is to give every single judge and staff attorney their own AI clerk,” he said.
He also said AI systems can be tested in ways human decision-makers cannot.
“I’m not saying all machines aren’t biased,” Klapper said. “I’m not saying my machine isn’t even biased. I’m saying we can test it and people have tested it. And that is the benefit over humans.”
In Riverside County, emails obtained by CalMatters show Klapper proposed that the court begin with a routine civil motion and then expand quickly if the tool proved useful. He also suggested court officials identify motions and workflows that were especially burdensome, mentioning the Racial Justice Act as one example.
About two weeks later, Riverside court officials identified labor-intensive areas such as discovery motions and attorney fee motions. For criminal cases, they suggested focusing on matters with large records, including death penalty habeas petitions and parole revocation.
Jason Galkin, Riverside County Superior Court’s chief executive officer, said the court is still in an exploratory stage.
“We don’t even know if expansion is likely,” Galkin said, adding that the immediate question is whether the tool helps staff do their jobs.
Across California, most superior courts now have policies governing generative AI, according to public records obtained by CalMatters. About a dozen of the 51 courts that responded said they are using AI-powered tools from companies such as LexisNexis, Thomson Reuters and Microsoft.
Los Angeles County Superior Court Judge Samantha Jessner, who chairs the Judicial Technology Advisory Committee, said she only recently became aware that the Learned Hand tool could eventually be used outside civil court. Judges are generally not involved in contract negotiations because of ethical limitations, she said.
Jessner said the judiciary has a responsibility to examine whether AI has a useful role in the courts, while also proceeding cautiously.
Hochman said he expects AI will become part of the legal system but believes it is better suited for repetitive, low-level work than for analysis that affects the outcome of a case.
“The only thing I’m 100% sure of is that AI didn’t go to law school,” he said.
Original source: CalMatters
