From the Citizen Engagement & Services edition of October 30, 2026
California's Judicial Council cancels a statewide AI translation rollout at court counters after advocates document translation errors from the pilot
Judicial Council of California; San Diego Superior Court; Legal Aid Foundation of Los Angeles; California Rural Legal Assistance · State court public counters (language access) · California, United States
- Publisher
- Times of San Diego
- Original publication
- 2026-10-24
- Source retrieved
- 2026-10-31
- Event date
- 2026-10-15
What happened
The Judicial Council issued an RFP in late August for statewide AI voice-to-text translation at public counters, with a Dec. 1 contract start, then published a cancellation notice Oct. 15 and plans an internal review of accuracy, efficiency and effectiveness. The pilot ran in more than 30 courts from 2021 to April 2026. San Diego used it about once a month, and clerks preferred a phone interpreter service; staff said it struggled with fast speakers, needed updates and lacked Mandarin and Haitian. Legal aid groups obtained transcripts through records requests; professional translators found errors such as DUI translated as IUD and 'no contest' as 'no competition.' The council has not published its preliminary evaluation.
Why it matters
Language access at public service counters is core to this stream; the courts setting overlaps the public-safety stream, but the lesson applies to any agency front desk using AI interpretation. It shows that a multi-year pilot can still fail on accuracy for high-stakes interactions and that records requests can surface the evidence.
Evidence and measured results
Reporting based on public records obtained by legal aid groups, court spokesperson statements and a professor's comment. Error examples come from advocates' review; transcripts are not tied to specific courts and the council's own evaluation is unpublished.
Limitations and uncertainty
Advocate-compiled error samples; no error rate; official evaluation not public; cancellation reason given officially as evolving technology.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-31; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Agencies want faster language help at counters, but AI translation errors can harm people in high-stakes interactions.
- Stakeholders
- Court or agency language access coordinator
- Court executive officer or agency director
- Legal aid and community partners
- Procurement lead
- Discovery questions
- Which counter interactions are rights-affecting and must go to a human interpreter?
- Has any current pilot been scored by professional translators?
- What fallback do staff use today, and how often?
- Value hypothesis
- Independent evaluation could prevent a costly rollout of tools that do not meet accuracy needs; no savings are measured.
- Potential engagement
- An accuracy evaluation of an existing AI interpretation pilot using professional reviewers.
- Unsupported claims
- Any error rate for the pilot
- That human interpreters can be replaced
- Vendor-specific conclusions
Pre-sales engineering
Role takeaway
- Fit
- Applies to voice-to-text or speech translation at service counters and front desks.
- Architecture and integration
- Run alongside existing phone interpreter services, with a one-step handoff and transcript logging under a retention policy.
- Prerequisites
- Real counter audio for testing
- Professional translator reviewers
- Interpreter service contract
- Constraints
- Noise and fast speech at counters
- Uneven language coverage
- Legal terminology
- Security
- Transcripts contain sensitive personal and legal details; define storage, access and records-release handling.
- Proof of value
- Score output by language on real audio with professional reviewers
- Measure handoff rate to human interpreters
- Test legal terms such as plea and protective-order vocabulary
Delivery
Role takeaway
- Work
- Evaluate and, if used at all, bound AI interpretation to low-stakes counter questions.
- Dependencies
- Interpreter services vendor
- Language access coordinator
- Legal aid and community review
- Ownership
- Court or agency language access office.
- Skills and adoption
- Clerks need training on when to stop AI use and switch to a human interpreter.
- Governance checkpoints
- Published evaluation before any scale-up
- Community and legal aid review
- Transcript retention review
- Acceptance criteria
- Proposed: per-language accuracy threshold set by professional reviewers before use
- Proposed: 100% of rights-affecting interactions routed to human interpreters
- Risks
- Mistranslation of legal instructions
- Loss of public trust
- Procurement started before evaluation
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Voice-to-text translation at a noisy counter faces audio capture, fast speech and missing languages; a human phone interpreter fallback stayed in place throughout.
Governance
Who approves, reviews and stays accountable for outcomes?
High-stakes counter interactions need human interpreters or qualified review; public pilots should publish evaluation results before scaling.
Security and privacy
What data, permissions and controls need testing?
Transcripts of court-user conversations were retained and later released through records requests, a reminder to define retention and redaction for AI interaction logs.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Errors fall on limited-English court users seeking restraining orders or handling fines; clerks preferred human phone interpreters.
Procurement
What should contracts, pricing and exit terms secure?
An RFP was cancelled after proposals arrived, showing that procurement should follow, not precede, a published pilot evaluation.
Operating model
Which teams own the service once it runs?
Keep human interpretation as the default for rights-affecting interactions; treat AI as limited support.
Publication history
Read preserved resource versions (JSON)Stable resource ID: california-courts-cancel-ai-counter-translation-rfp-20261024