California Judges Are Testing a New AI Clerk, and You Won’t Know if It’s Looking at Your Case

Why It Matters

If AI-generated analysis reaches criminal or Racial Justice Act petitions, errors could affect freedom and public confidence in the judiciary, while successful automation could alleviate massive case backlogs.

Key Takeaways

• Los Angeles court’s AI pilot costs about $314,000
• Riverside’s agreement is a $10,000 civil‑focused test
• Tool combines Anthropic, OpenAI and Google language models
• Contracts permit future use on criminal suppression motions
• Judges fear AI could erode confidence in high‑stakes rulings

Pulse Analysis

The pilot programs in Los Angeles and Riverside represent the first large‑scale experiments with generative AI as a judicial clerk. Learned Hand’s platform pulls from leading large‑language‑model providers to produce research memos, draft orders and tentative rulings, aiming to reduce the chronic backlog that plagues California’s superior courts. By automating routine legal research, the technology promises faster turnaround times for civil motions, freeing human clerks for more nuanced analysis. However, the contracts explicitly allow expansion into criminal divisions, where decisions on motions to suppress evidence or post‑conviction relief can determine a person’s liberty.

Critics warn that AI hallucinations and embedded biases pose a grave risk in high‑stakes contexts. Recent incidents—including a $10,000 fine for citing non‑existent cases and documented hallucinations in hundreds of California filings—highlight the technology’s propensity for factual errors. The Racial Justice Act, which lets incarcerated individuals challenge convictions on bias grounds, is especially sensitive; any AI‑generated recommendation could inadvertently reinforce systemic disparities. Judges and public defenders have voiced concerns that AI cannot grasp the nuanced social dynamics essential to fair adjudication, and that undisclosed use may undermine due‑process protections.

The broader implication is a looming policy crossroads for the U.S. judiciary. As courts grapple with mounting caseloads, AI offers a tempting efficiency boost, yet the lack of transparent evaluation metrics and mandatory disclosure standards hampers trust. State Judicial Councils are beginning to require AI use policies, but the effectiveness of bias‑testing regimes remains unproven. The outcome of these California pilots will likely shape national discourse on whether generative AI can be responsibly integrated into the justice system without compromising accuracy, equity, or public confidence.