AI Trends

Trending story

Stanford researchers find rising signs of AI-assisted drafting in court filings

1 reports1 story updatesUpdated Nov 20 00:25

Understand this story

AI overview

Researchers at the Stanford Institute for Human-Centered Artificial Intelligence and the Regulation, Evaluation, and Governance Lab found that signs of AI-assisted drafting in the Santa Clara County filings examined rose from nearly zero in 2022 to more than one-third in 2026. Their study created the PhantomDocket benchmark and proposed the ScoreMix calibration method; by 2026, AI usage in Santa Clara County non-debt-collection cases was estimated at 38%. However, detection rates for the same set of Santa Clara County documents varied from 10% to 98% depending on the settings, indicating that detectors are better suited to estimating overall prevalence than determining whether a specific document used AI. There are currently no other evidence-backed developments.

AI-generated from public reports

Latest development · Report

Stanford Study Tracks AI-Assisted Text in U.S. State Courts and Reveals Detection Challenges

Developments

1 confirmed developments

  1. Stanford researchers find rising signs of AI-assisted drafting in court filings

    Source:Stanford HAI News

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 20

  1. Stanford HAI NewsFirst-partySelected

    Stanford Study Tracks AI-Assisted Text in U.S. State Courts and Reveals Detection Challenges

    Stanford researchers developed the PhantomDocket benchmark and proposed the ScoreMix calibration method to track the use of generative AI in court filings across four U.S. jurisdictions. By 2026, estimated AI usage rates in non-debt-collection cases in Santa Clara and San Francisco were 38% and 36%, respectively, while those in Tulsa and Oklahoma were 17% and 16%, respectively. The study also found that detection rates for the same set of Santa Clara filings could vary from 10% to 98% depending on the settings, meaning detectors are better suited to estimating overall adoption rates than determining whether a specific filing used AI.

Related stories

  1. Related storiesEnglish translation is being prepared.
  2. Related storiesEnglish translation is being prepared.
  3. Related storiesOpenAI Shares New Data on How AI Is Changing Its Internal Research
  4. Related storiesOpenAI Proposes a Full-Lifecycle Safety Review Framework for Frontier Models