Italiano English

WHERE THE NEXT BUSINESS BEGINS

← Back to the magazine
Technologies becoming businesses

AI moves from lawyers to judges: who prepares the decision

Clio acquires Learned Hand, which develops tools for reading case files and preparing judges’ work. The new market also affects those waiting for a decision: promised speed, necessary checks and responsibilities that remain human.

FRONTIERA · 2026-10-05 · 5 MIN · 1084 WORDS
The Stanley Mosk Courthouse in Los Angeles, an archive photograph from 2005
FRONTIERA MAGAZINE.AI moves from lawyers to judges: who prepares the decision
Carol M. Highsmith / Library of Congress · public domain, no known restrictions; photograph from 2005, resized and converted to WebP Source · Public domain; no known restrictions

A career moves to the other side of the table

Bridget Mary McCormack led the Michigan Supreme Court. Now she is joining Clio to build a commercial business serving the judiciary. Her appointment, announced in September with a start scheduled for October, puts someone who knows how courts work in charge of a market still being developed. On September 30, 2026, the industrial step follows: Clio announces its acquisition of Learned Hand, a startup developing artificial intelligence tools for judges and their staff.

Her choice makes a transformation less visible than applications for writing letters or summarising contracts tangible. The software aims to enter the stage at which a dispute is studied by the person who will decide it. For someone involved in a case, it matters not only when the answer will arrive. It also matters which elements will be considered, which errors can be identified and how understandable the path to the decision will be.

The work before the signature

Learned Hand presents a product that organises case documents, helps identify disputed issues and prepares working materials, including memoranda and first drafts of orders. Its website describes citations linked to documents in the case record, available for checking, and text that the judge can edit and export. These are features stated by the provider: they are not an independent assessment of their accuracy.

The interesting step is preparation. A summary can help someone navigate many pages, but selection also changes what receives attention. A correct reference does not, by itself, guarantee that all relevant elements have been included. The responsibility to read, connect and assess remains distinct from the program’s ability to produce plausible text.

Consider the reader’s question without inventing a case: if a decisive document is summarised incorrectly, who notices? If the first draft offers a convincing but incomplete explanation, who returns to the record? These are questions about the checking process, not errors we attribute to this product. The service’s value will also depend on how easy it makes disagreeing with its own answer.

An empty courtroom in the Joseph P. Kinneary federal courthouse in Columbus
Courtroom in the Joseph P. Kinneary federal courthouse in Columbus, Ohio. Archive photograph from 2009: it illustrates the judicial setting, not the use of Learned Hand. Carol M. Highsmith / Library of Congress · Public domain; no known restrictions

Klapper knows the room

Shlomo Klapper, Learned Hand’s founder, clerked for Judge Steven J. Menashi in the federal Second Circuit Court of Appeals. He also practised at Quinn Emanuel and worked at Palantir. His career connects judicial practice and technology development. The acquisition brings that experience into a company seeking to extend its reach from law firms to courts.

LawSites reconstructs a precise sequence: first the arrival of Pablo Arredondo, Casetext’s cofounder, to develop the product; then McCormack’s appointment for the new business; finally the purchase of a startup with an existing service. This is a different choice from beginning solely with a research promise. The financial terms of the acquisition have not been disclosed, as Reuters also reports.

For the service’s builders, technical expertise meets a particular test: the customer buys a tool, but its effects also reach the parties to proceedings. These people do not necessarily choose the provider. A court can assess time and organisation; those waiting for justice need improvements to be recognisable in the quality of examination as well, not simply in the number of pages produced.

The Michigan Hall of Justice in Lansing
Michigan Hall of Justice in Lansing, home of the Supreme Court formerly led by McCormack. Archive photograph from 2023. Tony Webster · CC BY 2.0

Los Angeles, with a necessary distinction

Clio lists the Superior Court of Los Angeles County among the courts with which Learned Hand works. LawSites recalls a collaboration announced in March to explore using AI in judicial work. This documents a relationship and experimentation; it does not justify reporting that every courtroom already uses the system or that decisions have been entrusted to a machine.

The sources consulted do not offer an independent comparison of case durations before and after adoption. Nor do they allow us to calculate how many errors the program removes or introduces. The acquisition is a fact. Shorter waits and greater working capacity are outcomes still to be verified. Keeping these levels separate makes it possible to follow development without having to choose between automatic enthusiasm and pre-emptive rejection.

Useful verification will require information from practice: which tasks are delegated, how long checking the outputs takes and how corrections are recorded. A minute saved in drafting is not necessarily a minute saved in proceedings. And a faster draft does not make discussion between the parties unnecessary. The change will need to be measured across the whole process.

California has already written a caution

There is a public document against which to assess the promises. Standard 10.80 of the California Standards of Judicial Administration, effective September 1, 2025, provides guidance on judicial officers’ use of generative AI in adjudicative activities. It is a California reference, not a universal rule or an Italian law.

The text recommends reasonable checks of accuracy and the correction of errors or invented information, including when materials are prepared by others. It addresses protecting confidential information when using public tools, discrimination and removing biased or harmful content. It also invites consideration of whether to disclose AI use in content made available to the public.

The distinction between a public system and a court-managed environment matters. So does the fact that this guidance certifies no provider. For readers, the document makes a practical question visible: which steps allow a human to check what arrives on the desk? The presence of checks carries more weight than simply calling a tool an “assistant”.

A market that must earn trust

An empty courtroom in the John W. McCormack Courthouse in Boston
Courtroom in the John W. McCormack Courthouse in Boston, archive photograph from 2009. A contextual image, not a documented trial of the product. Carol M. Highsmith / Library of Congress · Public domain; no known restrictions

The new business sits between a useful possibility and a responsibility that is difficult to transfer. Reducing preparation burdens could free attention for genuinely disputed points. It could also shift time from writing to checking. For judges’ staff, the required skills would change: knowing how to formulate a request to the system, recognise omissions and trace a source becomes part of the job.

This is an interpretation of possible consequences, not a forecast of jobs eliminated. The sources examined contain no documented plan that would allow that employment change to be quantified. The product opens organisational choices that will depend on the courts, the types of disputes and the checks adopted. Neither all case files nor all offices pose the same problems.

For Frontiera, the story begins when a technology meets people who must live with its results. Here the future takes shape in a preparatory step: the way a case becomes readable. The next test will be being able to see who checked a citation, who corrected a summary and which limitations remain. Someone waiting for a decision does not need a machine that appears confident. They need a process in which that confidence can be scrutinised.