Calibrated AI detectors estimate rising AI-assisted drafting in four state courts
Source check: Source checked. Not independent scientific confirmation.
Finding: Stanford researchers applied the learned EditLens detector with ScoreMix calibration to 127,518 complaints from four U.S. county court systems. Their selected settings estimated rising AI-like drafting, reaching about 38% in Santa Clara and 36% in San Francisco in 2026 for non-debt-collection complaints. Calibration used documented hallucination filings and assumed their detector-score distribution represented other AI-assisted writing. Detector settings, imperfect controls, case mix and selection of conspicuous AI mistakes limit the estimates. These are population estimates, not proof that any particular filing used AI, fabricated citations or affected a judicial outcome. November 19 is the institutional announcement date; the working paper posting day is unestablished.
Technology: EditLens (Llama-3.2-3B); ScoreMix calibration
Family Llama: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Learned text classifier supplies scores for calibrated population estimates.
- Test and result
- Retrospective court complaints and detector calibration/validation.
- Limitation / still unknown
- Exact prevalence depends on detector specifications and reference distributions.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://hai.stanford.edu/news/ai-might-be-flooding-state-court-houses-but-proving-it-is-another-matter · https://dho.stanford.edu/wp-content/uploads/GenAI_Courts.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review date: 20 Nov 2026. Review scope: Dated first-party announcement and bounded primary working-paper Methods Sections 2–5 and discussion. Separate appendix not reviewed; no individual-file attribution or independent replication..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.