route-fifty.com via Reddit

California Courts Deploy AI Clerk, Skip Disclosure

Key insights

  • LA County signed a $314,000 contract for Learned Hand, with expansion planned into criminal, family, and probate court divisions.
  • Current California rules only require AI disclosure when an entire document is AI-written, leaving partial AI drafting unreported to litigants.
  • Judges inside the pilot have already raised due-process concerns about the opacity of AI assistance in active cases.

Why this matters

Judicial AI tools operating without mandatory disclosure create a structural accountability gap where litigants cannot contest AI-influenced reasoning they never knew existed. The LA County contract's roadmap into criminal and family law means people facing incarceration or custody loss will soon be affected, raising the stakes well beyond civil disputes. For AI practitioners building enterprise tools in high-stakes government settings, this case signals that disclosure norms and procurement contracts are diverging in ways that will invite regulatory and constitutional challenge.

Summary

Two California superior courts are quietly using an AI tool called Learned Hand to draft judicial orders and research memos, with no obligation to tell the people whose cases it touches. Los Angeles and Riverside County Superior Courts are running the pilot, with LA County holding a $314,000 contract slated to expand into criminal, family, and probate divisions. The tool runs on a combination of Anthropic, OpenAI, and Google models, none of which are disclosed to litigants by default. Essentially: (Anthropic, OpenAI, Google) are powering undisclosed AI assistance inside active court proceedings. - Current California disclosure rules only trigger when AI writes an entire document, leaving partial AI drafting completely unreported. - Judges inside the pilot have already raised due-process concerns about the opacity. - The $314,000 LA County contract signals this is institutional infrastructure, not a small experiment. The disclosure gap here is not hypothetical: litigants in criminal and family cases could have outcomes shaped by AI-drafted orders without ever knowing it.

Potential risks and opportunities

Risks

  • Litigants in already-adjudicated LA or Riverside County cases could seek reversal if AI involvement is retroactively disclosed, generating an appeals backlog that undermines the efficiency rationale for the tool
  • Anthropic, OpenAI, and Google face reputational and contractual exposure if undisclosed AI involvement surfaces in a high-profile criminal or custody ruling in the next 12 months
  • California's Judicial Council faces pressure to issue emergency disclosure guidance before the criminal division rollout, or risk federal due-process challenges that could void pilot-era rulings statewide

Opportunities

  • Legal AI vendors building auditing or explainability layers (Harvey, Casetext/Thomson Reuters) can position their products as disclosure-compliance infrastructure for court systems procuring AI clerks
  • AI governance consultancies and civil procedure law firms could see demand from litigants seeking discovery into AI's role in past decisions, creating a new litigation vertical
  • States watching California can differentiate their own AI clerk procurement by mandating disclosure from the start, creating a compliance market for auditable judicial AI and pressure on vendors to build disclosure tooling natively

What we don't know yet

  • Whether any litigant has already had a final order drafted or influenced by Learned Hand without disclosure, and whether that creates grounds for appeal under existing California procedure
  • How Anthropic, OpenAI, and Google have structured liability exposure in the LA County contract, given that due-process concerns have already been raised internally by pilot judges
  • What the specific timeline for criminal and probate division rollout looks like, and whether California's Judicial Council will update disclosure rules before that expansion begins