Courts across the United States are facing mounting pressure as caseloads continue to grow without a corresponding increase in resources. The Los Angeles Superior Court has launched a pilot program to test whether artificial intelligence can ease that burden by handling routine administrative work. The program centers on an AI tool called Learned Hand, which summarizes filings, organizes evidence, and generates draft rulings in civil cases. The goal is to free judges from time-consuming clerical tasks so they can concentrate on legal analysis and discretion.
Shlomo Klapper, founder and CEO of Learned Hand, said courts are under severe strain as caseloads rise while additional support remains unavailable. He noted that advances in AI are also contributing to the problem by making it easier and cheaper to produce legal filings. According to a February 2026 report by national law firm Fisher Phillips, filings increased 49 percent over the past year, climbing from 4,100 to 6,400. Klapper described this dynamic as AI simultaneously creating and potentially helping to solve a growing court backlog.
The pilot gives a select group of judicial officers access to the Learned Hand system across the full arc of a case, from intake through to draft rulings. Klapper, a former judicial law clerk for the U.S. Court of Appeals and a former deployment strategist at Palantir, founded the company in 2024. The tool is named after a federal judge of the same name and was designed specifically for courts rather than adapted from general-purpose AI systems. Klapper said the system is built to surface key facts and legal issues while leaving all judgment and agency with the presiding judge.
Presiding Judge Sergio C. Tapia II emphasized that the technology would not interfere with the independence of judicial decision-making. In a statement, he said the court is carefully evaluating emerging technologies to determine how they might help judicial officers work more efficiently and effectively. He was explicit that while the tool may change how judges review case files and information, it will not replace or compromise the impartiality of the judicial process. The court views the pilot as a measured and cautious step rather than a wholesale adoption of AI.
A central challenge in developing AI for legal settings is ensuring the accuracy of its outputs. Klapper said the most significant cost in running the system’s large language model comes not from generating text but from verifying that the output accurately reflects the underlying case materials and legal sources. He described generation as straightforward, while verification requires considerably more effort and expense. This focus on reliability is a direct response to concerns about AI producing inaccurate or fabricated information.
High-profile incidents have already demonstrated the risks of unchecked AI in courtrooms. In 2023, the defense team for Prakazrel Michel, a founding member of hip-hop group the Fugees, alleged that an AI-assisted closing argument included frivolous claims and overlooked weaknesses in the government’s case. That same year, a federal judge ordered lawyers representing former Trump attorney Michael Cohen to submit printed copies of cited cases after the court was unable to verify them. These cases have heightened scrutiny around the use of AI-generated legal content.
To reduce the risk of similar errors, Learned Hand operates within a defined and narrow pool of legal materials rather than drawing from the open internet or broad datasets. Klapper noted that large language models can reflect biases present in their training data, and limiting the source material helps address that problem. The system also breaks tasks into discrete steps, with each step assigned to a model that performs a specific function. Judges do not require any technical training to use the tool, which Klapper described as entirely point-and-click with no need for manual prompting.
Klapper acknowledged that AI tools must earn trust through demonstrated reliability rather than assumption. He said judges should not accept AI outputs at face value and that both the tools and the companies behind them carry a responsibility to prove their worth. His stated principle is to verify rather than trust, and he said the system is designed with that standard in mind. The Los Angeles pilot is intended to generate the kind of real-world evidence needed to assess whether AI can responsibly support the judicial process.
Originally reported by Decrypt.
