The Algorithmic Gavel: California's Dangerous Gamble with AI in the Courtroom

A quiet but profound revolution is underway within the hallowed halls of California’s justice system. The Los Angeles and Riverside County Superior Courts have embarked on pilot programs employing an artificial intelligence tool named “Learned Hand” to draft orders, produce research memos, and assist judges. While initially focused on civil matters, contractual roadmaps explicitly permit testing this technology in the criminal, family, and probate divisions—arenas where the stakes are nothing less than personal liberty, family unity, and fundamental fairness. This move, framed as a solution for efficiency and backlog reduction, represents one of the most significant and perilous encroachments of automation into a domain that demands quintessentially human judgment. As a firm believer in democratic institutions, the rule of law, and the Bill of Rights, I view this development not with techno-optimism, but with deep-seated alarm for the integrity of our justice system.

The Facts: A Pilot Program with Expansive Ambitions

According to reports, the Los Angeles County Superior Court holds a contract worth approximately $314,000 with Learned Hand, a company founded by Shlomo Klapper, a former federal court clerk and Palantir employee. Riverside County has a separate $10,000 agreement. The tool aggregates language models from Anthropic, OpenAI, and Google to function as an “AI clerk.” In Riverside, civil and probate attorneys are using it to draft research memos. In Los Angeles, six judges and their research attorneys are using it for research, motion summaries, and assisting in drafting tentative rulings in the civil division.

The immediate controversy stems from the contract’s scope. It allows for the tool’s use on two critical criminal motions: motions to suppress evidence and motions for post-conviction relief. Furthermore, during internal presentations, Superior Court Judges Yvette Verastegui and Olivia Rosales suggested the tool could eventually assist with petitions filed under California’s 2020 Racial Justice Act, which allows incarcerated individuals to challenge convictions or sentences they believe were tainted by racial bias. This prospect triggered immediate concern from an anonymous Los Angeles judge, who called the idea “outrageous,” warning it would “erode the public’s confidence.”

Court executives, including Los Angeles County Superior Court Executive Officer David Slayton and Riverside CEO Jason Galkin, emphasize a cautious, evaluative approach. They state expansion beyond civil cases will not occur until leadership is “comfortable” and the tool is proven effective. Klapper argues the judiciary desperately needs AI to manage workloads, stating the only solution is “to give every single judge and staff attorney their own AI clerk.” He acknowledges potential bias but asserts it can be tested—a benefit over humans.

The Context: A Pattern of AI Failures in Law

This experiment does not occur in a vacuum. As noted in the report, AI’s integration into legal processes has been marred by notorious failures. “Hallucinations”—where AI invents non-existent case law or facts—have already infiltrated the system. Researcher Damien Charlotin has documented hundreds of such errors, including nearly 90 in California courts since August 2024. A Los Angeles lawyer was fined $10,000 for citing fake cases, and Nevada County prosecutors made errors in four cases due to AI. A 2025 MIT study suggested models from major companies can reduce critical thinking.

Furthermore, the fundamental issue of bias is not theoretical. Large language models have a documented history of demonstrating race and gender bias. Analyses of predictive policing tech and risk-assessment algorithms like COMPAS have found they disproportionately label Black individuals as higher risk. This history casts a long, dark shadow over any proposal to use AI for evaluating Racial Justice Act petitions, which are, by definition, nuanced inquiries into systemic and implicit bias.

Opinion: Efficiency as a Trojan Horse for Erosion

The drive for efficiency is understandable. Court backlogs are real and justice delayed is justice denied. However, when that drive leads us to outsource core judicial functions to opaque algorithms, we have confused means with ends. The end of our justice system is not efficiency; it is justice itself—a profoundly human endeavor rooted in empathy, context, and moral reasoning.

The anonymous judge’s warning is prophetic: this path is “extremely perilous.” The danger is multifaceted. First, there is the risk of factual error and hallucination, magnified a thousandfold when applied to matters of liberty. Los Angeles County District Attorney Nathan Hochman articulated this perfectly: “I don’t want to take the chance… that AI happens to get it wrong. And now someone’s constitutional rights have been infringed.”

Second, and more insidiously, is the problem of bias replication and dehumanization. As Deputy Public Defender Elizabeth Lashley-Haynes, who specializes in Racial Justice Act cases, stated, these petitions are “incredibly nuanced” with “words that have racial undertones or racial meanings… way beyond the realm of anything that artificial intelligence could do.” Using an AI to generate “tentative decisions” on such petitions, as was suggested, is an act of profound disrespect to the lived experiences of those alleging bias. It reduces a deeply personal claim of systemic failure to a data point for processing, “reimpos[ing] a one-size-fits-all style of justice,” as the judge feared.

Third, this experiment threatens the transparency and accountability that are bedrock principles of our legal system. The current pilot use policies only require disclosure if a document is written entirely by AI, leaving a vast grey area of AI-assisted reasoning unexplained. The public, and the parties before the court, have a right to know when and how AI is shaping decisions that affect their lives. The courts’ refusal to confirm whether plaintiffs are aware their cases are part of this test is antithetical to open justice.

Klapper’s argument that AI bias can be “tested” while human bias cannot is a dangerous fallacy. Human judges are accountable. They can be appealed, their reasoning can be scrutinized, and they operate under a public code of ethics and canons of judicial conduct. An AI model’s “reasoning” is often an inscrutable black box, its training data proprietary, and its biases systemic and harder to pinpoint. Replacing one source of potential bias with an unaccountable, corporate-owned, and potentially more virulent one is not progress.

The Path Forward: Guardrails, Not Gavels

The appropriate role for AI in courts is not as a clerk or a junior judge, but as a highly constrained administrative tool. As DA Hochman suggested, it may be suited for low-level, repetitive, routine tasks that do not involve analytical judgment. Automating docket management, transcription, or legal research citation checking (with human verification) is a world apart from using it to analyze the merits of a motion that could set someone free.

Before any expansion occurs, several non-negotiable conditions must be met. First, a complete moratorium on using AI in any criminal, family, or juvenile matter until independent, peer-reviewed studies—not corporate white papers—prove extraordinary accuracy and a complete lack of discriminatory impact. Second, absolute transparency: any party whose case is involved in testing must provide informed consent, and any judicial work product influenced by AI must contain a prominent, detailed disclosure of its role. Third, all algorithms and training data used must be open to audit by an independent public body to root out bias.

We stand at a crossroads. Will we allow the imperative of efficiency and the allure of technology to erode the human foundation of our justice system? Or will we reaffirm that liberty, fairness, and dignity are values too sacred to be processed by an algorithm? The promise of a fair trial, of equal protection under the law, is the cornerstone of our social contract. We must not outsource its delivery to machines whose logic we cannot fully comprehend and whose priorities are not justice, but pattern recognition. The gavel must remain in human hands, for only human hands can weigh the gravity of liberty and the depth of injustice. Our courts must remain temples of reasoned humanity, not factories of automated output.