Six civil court judges in Los Angeles County have quietly begun piloting an artificial-intelligence system that drafts tentative rulings directly from case filings, a move that places the nation’s largest trial court at the center of a national debate over machine-generated justice.
The $300,000 program, launched under the radar this month and scheduled to run through early 2027, is among the first in the United States to embed generative AI inside a state courthouse’s day-to-day workflow. Each AI draft must still be reviewed, edited and formally adopted by a human judge before it becomes binding, court administrators emphasized, but the software already produces bench memoranda, statutory citations and proposed orders within minutes of a case being called.
“The goal is not to replace judicial discretion—it is to amplify it,” Presiding Judge Samantha R. Jessup said in a written statement. “If we can reduce the time a self-represented litigant waits for a ruling from twelve weeks to four, that is a measurable improvement in access to justice.”
Inside the Pilot
The pilot is limited to civil motions—evidentiary disputes, demurrers and discovery matters—heard in the Stanley Mosk Courthouse downtown. Judges volunteer to opt in; attorneys must affirmatively consent to have their papers fed into the large-language-model stack, which was fine-tuned on roughly 40,000 anonymized tentative rulings dating back to 2014. The system, supplied by San Francisco start-up Judicata, does not access criminal files or family-law cases involving minors.
Once a motion is filed, the platform ingests PDFs, extracts key legal issues and, referencing California code sections and controlling precedent, produces a five- to ten-page draft. Judges can accept, modify or discard the document. Early metrics show an average drafting time of seven minutes versus three hours for manually written tentatives, according to internal figures shared with The AI Era.
Ethical Guardrails
The rollout arrives amid heightened scrutiny of AI “hallucinations” in legal writing. A widely cited Wikipedia entry documents instances where attorneys were sanctioned for submitting briefs that cited non-existent cases generated by chatbots. Los Angeles County’s vendor says it has instituted a two-step verification layer: every statute or case citation is cross-checked against an encrypted, read-only copy of Thomson Reuters’ Westlaw database before the draft is released.
“We are not asking the model to be creative; we are asking it to be accurate,” said Judicata chief executive Clark H. Rowley. “If the confidence score on any proposition falls below 92 percent, the paragraph is flagged and the judge is warned.”
Judicial ethics experts remain cautious. “A tentative ruling is still a judicial communication,” said Professor Miriam Baer of Brooklyn Law School. “If an AI introduces subtle bias—say, by recommending harsher sanctions against pro se litigants—that bias can shape settlement negotiations even before a final order is entered.”
Broader Digital Shift
The county’s experiment parallels wider efforts to digitize courthouses nationwide. Twenty-one state legislatures now operate paperlessly on the NeVA platform, while policy makers urge the integration of AI and data analytics for stronger early-warning systems that flag docket bottlenecks before they cascade into backlogs.
Los Angeles County itself fields roughly 350,000 new civil cases annually. Court Executive Officer Sherri R. Carter estimates that if only one-third of eligible motions adopt the tool, the county could free up the equivalent of 1.8 judge-years annually—resources that could be redirected to trial-ready cases languishing beyond statutory time limits.
Attorney Reaction
At a bar association town-hall last week, litigators voiced mixed opinions. “Anything that speeds up tentative rulings is welcome,” said Jessica Wu, a partner at a midsize firm. “But I need assurance that the AI is not quietly favoring the moving party because training data historically favors denial of demurrers.”
Public-interest groups worry about transparency. The current pilot does not require disclosure when a ruling is AI-drafted, although judges must retain an audit log. UCLA’s David J. Epstein Program on Public Interest Law has petitioned the court to amend its rules so that any AI-generated tentative includes a docket notation. “Transparency is not optional when constitutional rights are at stake,” said clinical director Blanca S. Valencia.
Next Steps
An independent evaluation by the Judicial Council of California is slated for mid-2026. If success metrics—speed, accuracy, user satisfaction—meet benchmarks, the program could expand to probate and limited civil departments across the county, potentially influencing how other mega-jurisdictions such as Cook County, Illinois, and Harris County, Texas, approach AI adoption.
Meanwhile, Judicata is negotiating with the Administrative Office of the Courts for a parallel pilot in San Diego and San Francisco. The company, which raised a $17 million Series B last December, competes with Thomson Reuters’ CoCounsel and LexisNexis’ Protégé, both of which are marketing generative brief-drafting tools to private firms.
Global Context
Outside the United States, AI-assisted judging is no longer theoretical. Estonia’s Ministry of Justice has tested an AI that adjudicates small-contract disputes under €7,000, while Chinese internet courts deploy AI to transcribe testimony and recommend sentences in copyright infringement cases. The difference, scholars note, is that those systems operate within specialized tribunals where parties can opt for human review. Los Angeles County’s experiment embeds AI inside a general-jurisdiction court where the stakes—and the docket—are far larger.
For now, the judges say they will proceed deliberately. “We are not handing the gavel to a machine,” Judge Jessup reiterated. “We are using every available tool to deliver faster, fairer justice. If the tool fails, we will shelve it. If it succeeds, we will scale it—carefully.”
0 Comments