How AI Is Being Used to Clear Court Backlogs in LA

Courts astir the satellite are straining nether increasing caseloads, and a aviator programme successful Los Angeles is hoping to alteration that by investigating whether AI tin assistance judges without offloading their judgment.

The Los Angeles Superior Court is investigating an AI instrumentality called Learned Hand that summarizes filings, organizes evidence, and generates draught rulings successful civilian cases.

The extremity is to trim clip spent connected administrative tasks truthful judges tin absorption connected the parts of a lawsuit that necessitate ineligible investigation and discretion, Learned Hand laminitis and CEO Shlomo Klapper told Decrypt.

"We're astatine a spot successful nine wherever courts are nether tremendous strain," Klapper said. "Their caseloads spell up, but nary assistance is coming,” helium said, adding that advances successful artificial quality are “massively dropping the outgo of litigation."

AI is expanding unit connected the courts by making it easier to nutrient filings, with filings rising 49% from 4,100 to 6,400 successful the past year, according to a February 2026 report by nationalist instrumentality steadfast Fisher Phillips.

The Los Angeles Superior Court aviator gives a tiny radical of judicial officers entree to Learned Hand’s AI strategy to trial its show crossed a case, from intake to draught rulings.

A erstwhile judicial instrumentality clerk for the U.S. Court of Appeals and deployment strategist with Palantir, Klapper said Learned Hand, founded successful 2024, and named aft a national justice of the aforesaid name, was designed to springiness overburdened courts “purpose-built” AI tools that chopped down connected “drudge work” by surfacing cardinal facts and ineligible issues portion leaving judgement and bureau with the quality judge.

“With this partnership, we are cautiously evaluating emerging technologies to find however they whitethorn enactment judicial officers successful moving much efficiently and effectively,” Presiding Judge Sergio C. Tapia II said successful a statement. “Let maine beryllium clear—while this instrumentality whitethorn heighten the mode judicial officers reappraisal and prosecute with lawsuit files and information, it volition not replace, oregon successful immoderate mode compromise, the sanctity, independence, and impartiality of judicial decision-making.”

Klapper said the harder portion of processing an AI for courts is not generating substance but checking AI output against the underlying lawsuit materials and ineligible sources.

“Most of the disbursal of our ample connection exemplary is successful the verification, not the generation,” Klapper said. “Generation is easy. Anyone tin make something, but however bash you marque definite that it's truly reliable?”

AI hallucinations person already surfaced successful high-profile tribunal cases.

In 2023, the defence squad for Prakazrel “Pras” Michel, a founding subordinate of hip-hop radical the Fugees, alleged that an AI helped constitute a closing statement that included frivolous claims and missed weaknesses successful the government’s lawsuit against him.

That aforesaid year, a national justice ordered lawyers representing erstwhile Trump lawyer Michael Cohen to supply printed copies of cited cases aft the tribunal could not verify them.

Klapper said Learned Hand is built astir a narrower excavation of root worldly to trim the hazard of AI hallucinations. Rather than pulling from the unfastened net oregon random datasets, the strategy operates wrong a defined acceptable of ineligible materials.

The crushed is that ample connection models tin bespeak biases successful their grooming data, pointing to examples of AI echoing proposal from platforms similar Reddit, Klapper said. Learned Hand addresses that by breaking tasks into steps and assigning each measurement to a exemplary with a circumstantial function.

Learned Hand is besides designed truthful that judges bash not request method grooming to usage it.

"It's constituent and click," Klapper said. "They don't person to bash immoderate prompts."

Klapper argued that overmuch of a judge's time is spent connected regular tasks alternatively than ineligible reasoning, and that the AI aims to let them to “spend much clip connected justice enactment and little clip connected drudge work.”

Klapper said judges should not instrumentality AI outputs astatine look worth and that some the tools and the companies down them request to beryllium their reliability.

“I similar to say, don’t trust, verify,” helium said. “They shouldn’t spot anything. It has to amusement its worth.”