An AI tool is already drafting tentative judicial orders and research memos for judges in two of California's largest court systems — and neither Los Angeles County nor Riverside County is required to tell the people whose cases it touches.
The disclosure emerged from a CalMatters/KPBS investigation published May 26 by reporters Cayla Mihalovich and Khari Johnson. Court records obtained by the outlets show that the Los Angeles County Superior Court — the largest trial court in the United States — signed a roughly $314,000 contract with an artificial intelligence startup called Learned Hand, launching a pilot in February 2026. Riverside County followed with a $10,000 agreement covering civil and probate research memos. Together, the two pilots mark the most significant deployment of AI judicial assistance in California history.
Six Los Angeles civil court judges and their research attorneys are using Learned Hand to conduct legal research, summarize motions, and assist in drafting tentative rulings. The tool draws on large language models from Anthropic, OpenAI, and Google. The company says it has tested the system for bias and accuracy but has not published those results.
Both courts declined to say whether litigants are being told the tool is being tested in connection with their cases. A Los Angeles Superior Court spokesperson told CalMatters that testing is being conducted on motions that have already been decided, outside active case environments — but the contract itself allows testing on live cases.
LA County Contract Targets Expansion Into Criminal and Family Courts
The contract is not limited to civil proceedings. It includes a formal roadmap to expand Learned Hand's use into the criminal, family, and probate divisions — courts where the consequences of errors extend far beyond financial disputes and into people's freedom, custody of their children, and access to estate assets.
Court officials would not describe to CalMatters the criteria they are using to evaluate whether expansion into those divisions can proceed safely. Judge Samantha Jessner, who chairs the court's Judicial Technology Advisory Committee, said she was unaware until recently that the contract extended beyond civil cases. Jessner framed the pilot as an institutional obligation: "I think we have a duty and obligation to explore whether or not there is a place for artificial intelligence in what we do as a judicial branch."
The court's executive officer, David Slayton, said the tool will not move beyond the civil division until court leadership is satisfied with the results.
Judges Discussed Using AI to Review Racial Bias Appeals
The most sensitive detail in the investigation came from a judge inside the Los Angeles Superior Court who spoke on condition of anonymity because of judicial rules of conduct. That judge told CalMatters that colleagues at a recent luncheon had discussed using Learned Hand to help evaluate petitions filed under California's Racial Justice Act.
The Racial Justice Act — passed in 2020 and extended in stages to cover all felony convictions as of January 1, 2026 — allows prisoners to challenge convictions or sentences they believe were tainted by racial bias. Evaluating those petitions requires judges to weigh statistical patterns, witness credibility, historical discrimination, and systemic racism in prosecutorial charging decisions. It is among the most complex and high-stakes judgment tasks in the California court system.
"I think it is outrageous," the judge said. "AI cannot and never will be able to replace human judgment in evaluating complex social dynamics. Ultimately, that will erode the public's confidence in the competence and fairness of the judiciary."
Learned Hand CEO and founder Shlomo Klapper acknowledged the gravity of the question. "It's a very fraught decision, I'm not going to lie — extremely high stakes, a scenario where I understand people might be very concerned," he told CalMatters. He said any Racial Justice Act module would need to be developed in collaboration with the court and separately evaluated for bias before deployment.
What No Disclosure Rule Means for California Litigants
Neither court is currently obligated to inform litigants when AI assists in preparing orders or research memos connected to their cases. California's Judicial Council rules require disclosure only when a motion, decision, or other document is written entirely by generative AI. Partial AI assistance — which is what Learned Hand provides — triggers no disclosure obligation.
A survey conducted by Kolmogorov Law in April 2026 found that 73 percent of Californians were unaware that courts were using AI to help draft rulings, and 91 percent said disclosure should be required. No California rule and no federal rule currently guarantees a litigant the right to know whether AI participated in preparing a ruling in their case.
Los Angeles County District Attorney Nathan Hochman reviewed the contract and flagged two categories of criminal motions it permits: motions to suppress — which determine what evidence prosecutors may present at trial — and post-conviction relief motions filed by people already convicted. "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 said. "I don't want to take the chance, particularly in a criminal case, that AI happens to get it wrong."
How Does Learned Hand Actually Work in a Courtroom?
Klapper, who previously clerked for a federal appeals court and worked at surveillance technology firm Palantir before founding Learned Hand, describes the system as giving every judge their own AI clerk. The tool allows judges to upload case materials and receive structured preparation — record organization, motion summaries, legal research, analysis, and draft tentative rulings — tailored to the type of motion at hand.
The company evaluates its output against the standards applied to human law clerks: accurate legal research, sound analysis, and neutral, judge-ready writing. Klapper argues that AI's testability gives it an advantage over human clerks. "I'm not saying all machines aren't biased," he told CalMatters. "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."
A majority of California's 58 superior courts now have generative AI use policies — a condition the Judicial Council imposed before any court could adopt the technology. About a dozen have deployed AI tools from LexisNexis, Thomson Reuters, and Microsoft Copilot. But the scale and scope of Learned Hand's deployment in Los Angeles appears to go further than most of those implementations, and its criminal-expansion roadmap is unique among the courts that responded to CalMatters' public records requests.
AI Hallucinations Have Already Reached California Courts
The concern about error is not hypothetical. Researcher Damien Charlotin, a fellow at HEC Paris's Smart Law Hub who maintains a global database of AI hallucination incidents in legal proceedings, had catalogued nearly 90 such cases in California state and federal courts alone since August 2024 — part of a global tally exceeding 1,200 documented cases as of early 2026. The incidents range from fabricated case citations to invented statutes and misrepresented holdings.
In September 2025, an LA-based attorney received a $10,000 fine for submitting AI-generated case citations that did not exist. More recently, the Sacramento Bee reported that AI contributed to errors in four cases handled by prosecutors in Nevada County. Earlier this year, the U.S. Senate investigated federal judges in Mississippi and New Jersey for using generative AI to draft rulings with serious factual errors.
Most of these incidents involved attorneys or self-represented litigants. But UCLA Law School professors have warned that more judges will make AI-fueled errors as the technology spreads to judicial chambers.
A 2025 MIT Media Lab study found that participants who used ChatGPT to write essays showed lower brain engagement, weaker memory recall, and reduced critical thinking compared to those who used traditional search or wrote unaided. The researchers described the pattern as a form of "cognitive debt" — short-term convenience at the cost of long-term reasoning capacity. The study was conducted in an educational context and remains a preprint awaiting full peer review, but its implications for a profession whose core value is independent legal reasoning have drawn attention among legal scholars and judicial ethics researchers.
Frequently Asked Questions
Are California courts required to tell litigants when AI is used in their case?
Currently, no. California Judicial Council rules require disclosure only when a document is written entirely by generative AI. Partial AI assistance in research or drafting — which is how Learned Hand is deployed — carries no mandatory disclosure obligation. No federal rule covers the gap either, meaning litigants in California courts currently have no guaranteed right to know AI participated in preparing a ruling in their case.
What is Learned Hand and how is it being used in Los Angeles courts?
Learned Hand is a judicial AI company whose software draws on large language models from Anthropic, OpenAI, and Google. In Los Angeles, six civil court judges are using it to summarize motions, conduct legal research, and assist in drafting tentative rulings. The company has tested the tool for bias and accuracy but has not published those results publicly.
Can AI be used to draft court orders in California?
Yes, under current rules. The California Judicial Council's 2025 guidelines permit AI-assisted drafting as long as a judicial officer reviews and approves the final work product. The Los Angeles contract includes a roadmap to expand use from civil cases into criminal, family, and probate divisions — areas where errors can affect a person's freedom or custody of their children.
What are the risks of AI in the courtroom?
The documented risks include hallucinated case citations, invented legal standards, and automation bias — the tendency for human reviewers to over-rely on AI output without adequate independent verification. As of early 2026, researcher Damien Charlotin had catalogued more than 1,200 AI hallucination incidents in courts globally, with nearly 90 in California alone since mid-2024.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
