California Courts AI Translation Dui Iud No Contest Cancelled
I went to a California court counter for help with a restraining order, and the interpreter that day was an AI voice-to-text tool. It translated the clerk's instruction as: write down the name of the person "who receives the phone number, and address of the restraining order" — the clerk had actually said "who's getting the restraining order." In other pilot transcripts the same tool rendered "DUI" as "IUD" and "no contest" as "no competition." It didn't stumble or ask again, and it can't clarify who "he/she/they" means — it picks one subject and keeps going. Is that really what automated translation should mean when someone's legal rights are on the line?
1 Answer
Expert: Joann Lee, Special Counsel, Legal Aid Foundation of Los Angeles Verdict: the AI translation was wrong — and wrong in the way that matters most. It did not fumble a rare word; it inverted the legal meaning of ordinary ones. What the pilot transcripts actually show, according to professional translators who reviewed hundreds of pages obtained by legal aid groups: - "DUI" (driving under the influence, a criminal charge) came out as "IUD" (a contraceptive device). - "No contest" — a plea — came out as "no competition," which sounds like a sporting event. - A restraining-order instruction flipped its subject: the clerk asked for the name of the person who is *getting* the restraining order, and the tool produced a request for the name of the person who "receives the phone number, and address" of that order. - "Fear that he/she/they could do something to me" became "fear that you do something to me." The tool could not resolve the subject — and did not say so. That last point is the core defect. A machine translation failure is fluent and categorical: it produces a complete, confident sentence with no marker of where the guess happened, so the person at the counter has no way to know the sentence was rewritten. Joann Lee of the Legal Aid Foundation of Los Angeles draws the contrast with human interpreters: they may leave out details, but they rarely deliver the opposite meaning. "You might be giving really incorrect and harmful information through the machine," she said. Norma Ventura of California Rural Legal Assistance adds that a literal rendering "might be correct but contextually is inappropriate and renders the whole sentence incorrect." The record around the rollout is worth reading too. More than 30 California courts piloted the voice-to-text tool; the programme ran from 2021 to April 2026, and San Diego Superior Court used it about once a month in its Civil Business Office from 2022 until this spring. Court staff had already reported the failure modes in their own surveys — difficulty capturing fast speech, background noise interfering with translation, inconsistent quality across languages, and no support for Mandarin or Haitian. The Judicial Council issued a request for proposals in late August for a statewide rollout with a contract starting 1 December, then published a notice of cancellation on 15 October — the day before its 16 October meeting, where advocates presented the transcript findings. A spokesperson cited "evolving technology"; the council says it will run an internal review of "accuracy, efficiency, and effectiveness" first, and its preliminary evaluation of the pilot has not been made public. What correct looks like: every machine translation system makes mistakes, as University of Maryland computer scientist Marine Carpuat notes — the point is to design for that rather than pretend it away. At a clerk's counter, where a mistranslation determines which form a person files or what plea they enter, that means qualified human review at the point of delivery and certified interpreters as the default, not a fallback for when the software lacks a language. The pilot did not fail because the model was too small. It failed because nothing in the workflow was designed to catch a fluent wrong answer. Source: https://timesofsandiego.com/tech/2026/10/24/ai-translation-california-courts-pilot-project-rfp/
Your answer
Sign in to verify this AI response.
Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.
More in Law
- Fair Work Commission AI Fake Legal Precedents Riley Nuvei — Unidentified 'legally trained' AI tool
- AI Court Transcript Judge Bailiff Attribution Errors Indiana — Generative AI transcription tool
- Claude Enterprise Inaccurate Citations Pro Hac Vice Sanction — Claude (Anthropic)
- Flycatcher Affable Failla AI Fake Citations 38877 Sanction — Unidentified AI tool
- Krafton ChatGPT Project X Subnautica Earnout Court Ruling — ChatGPT (OpenAI)
- Claude Haiku Philadelphia Police Fake Homicide Tip — Claude (Anthropic)
- ChatGPT Fabricated Witnesses New Mexico Murder Appeal — ChatGPT (OpenAI)
- Sixth Circuit Whiting Fake Citations Sanctions — Unidentified generative AI tool
- Campos Munoz Dog Custody Fake AI Case Citations California — Unidentified AI legal research tools
- Scott V Illinois ChatGPT Fabricated Citations 15000 Sanction — ChatGPT (OpenAI)
More from this topic
I was representing myself in an unfair dismissal claim, so I used an AI tool billed as "legally trained" to prepare my submissions. It handed me case law to cite as precedent — specific legal principles and authorities — and it stated them as settled law, no hedging at all. At the hearing the Fair Work Commission said those authorities "may have been AI hallucinations and lacked any legal basis." How was I supposed to know the citations were not real?
I ran a trial recording through an AI transcription tool and it handed back a clean, properly formatted transcript — every speaker labelled, nothing flagged as unclear or uncertain. It went in as the certified record. Then the Indiana Court of Appeals read it and found that a motion made by the prosecutor had been attributed to the trial judge, and an objection by the defendant had been attributed to the bailiff. The tool stated all of it as plain fact, with no hint that any speaker was a guess. Does a transcript like that still count as a true record on appeal?
I had five hours to file a 20-page opposition brief, so I used AI-assisted legal research on what I was told was the highest-grade enterprise version of Claude and filed the brief it drafted. It gave me a citation for every argument, in normal-looking case format, and never once suggested that any citation needed to be checked — I genuinely believed it had not made up cases. Opposing counsel read the brief and immediately moved for sanctions over inaccurate citations. Is that on the AI, or on me?