AI in L.A. Courts: A Flow Analysis of Backlog Relief vs. Systemic Risk

The operational crisis AI is being deployed to solve is not theoretical; it is a massive, quantified flow problem. At its core is a legacy error of staggering scale: roughly 464,000 criminal cases dating back to the 1980s had their arrest disposition reports never filed with the state Department of Justice. This single backlog affects the official records of more than 400,000 people, creating a systemic vulnerability in the justice system's data flow.

This is just one dimension of a broader crisis. The Los Angeles Superior Court itself handles a colossal annual volume, managing 1.3 million filings a year across 36 courthouses. The sheer throughput of new cases creates constant pressure, making it difficult to maintain accurate records and clear case pathways. The system is perpetually trying to catch up with new filings while also grappling with decades-old errors.

The scale becomes truly systemic when viewed through the lens of the federal Immigration Court. As of late 2025, that system carried a backlog of 3,377,998 active cases. This is a flow problem of a different magnitude, where the rate of new case intake (over 130,000 in a single fiscal year) is far outpaced by the rate of resolution. The result is a mountain of pending matters that AI is now being asked to help triage.

The thesis is clear: AI is a targeted response to a crisis defined by massive, quantifiable backlogs. Yet its impact is inherently limited by the sheer volume of legacy errors and the relentless flow of new filings. It can accelerate processing, but it cannot instantly erase decades of unreported data or instantly resolve a multi-million-case federal backlog.

AI as a Triage Tool: Measuring the Direct Flow Impact

The court's immediate AI intervention is a targeted pilot focused on accelerating the most time-intensive judicial tasks. A select group of L.A. civil court judges now use the Learned Hand AI tool to summarize motions and draft tentative rulings. This is not a full automation of decisions; the system is designed as a "co-intelligence" or "judicial sous chef," with judges required to review and edit all outputs before adoption. The core flow impact here is a direct reduction in the time spent on case preparation, a critical bottleneck in a shorthanded system.

Beyond direct judicial drafting, the court is implementing a major alternative flow to clear its docket. Presiding Judge Sergio C. Tapia II announced the court expects to route 3,000 civil unlimited cases to mediation this year. This is a deliberate shift of volume away from the traditional adjudication pipeline. By moving these cases to a settlement-focused process, the court aims to resolve a significant number of matters without a full trial, directly reducing the backlog of cases awaiting a judge's final decision.

Finally, AI is being deployed for operational tasks to free up human capital. Tools are being used in clerks' offices to streamline document processing, a function where AI has already demonstrated high accuracy. Evidence from other jurisdictions shows AI-driven systems can achieve 98% to 99% accuracy in tagging and indexing filings, equivalent to the workload of nearly 20 human employees. This automation redirects staff from repetitive, error-prone work toward more complex, human-centered tasks, improving overall court efficiency.

Guardrails and Unintended Flows: The Risks to Trust and Equity

The court's AI pilot is built on a clear guardrail: human review is mandatory. Judges using the Learned Hand tool are required to review and edit all drafts before adoption. This principle is now codified in California's new Rule 10.430, which mandates disclosure, human verification, and bias audits for any judicial automation. The framework is designed to reinforce, not replace, human judgment, treating AI as a "co-intelligence" or "judicial sous chef."

Yet public trust remains a fragile asset. While a majority of Americans see benefits in AI for case organization and faster resolutions, they draw a hard line at high-stakes judgment. The survey shows only 10% back AI for sentencing, with 75% demanding disclosure of any AI use. This creates a tension: the public supports operational efficiency but insists on human oversight for outcomes that define liberty and livelihood. Any perceived failure in the AI-assisted process could rapidly erode this hard-won trust.

The most significant unintended flow is already in motion. The court's own system error means nearly 330,000 people will have their criminal histories updated after decades of being unreported. This isn't a one-time data fix; it's a massive, systemic flow of new information that could trigger cascading consequences. Affected individuals may face job losses, professional license revocations, or firearm confiscations as employers and licensing boards act on the newly disclosed records. The court itself acknowledges the unknown impact on how these agencies will treat the new information, highlighting a secondary social and economic cost that AI cannot mitigate.