Observed arrival · 2026-11-11
Can the Other Side Steer Your AI Lawyer?
A working-paper site reports four experiments testing whether opposing-party framing can steer AI legal review across 17 models.
- For
- Researchers studying AI-assisted legal review
- Worth noticing
- The study compares each run with the same model’s clean control and reports Holm-adjusted significance tests.
Field notes
The experiments use fictional legal scenarios, including a share sale with 16 negotiable issues, and compare results with each model’s own clean control. The page reports about 12,000 assessments and distinguishes CLI-routed models from models tested through plain APIs. Its results include uncertainty intervals and Holm-adjusted significance tests. The author labels the paper unreviewed and invites corrections, so the findings are presented as preliminary rather than settled.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue