SSRN · July 14, 2026
Predicting Supreme Court of the United States Outcomes with Frontier Models
An empirical study of how frontier AI models perform when predicting U.S. Supreme Court outcomes, with attention to accuracy, consistency, and the limits of model-generated legal reasoning.
Summary
This paper tests frontier AI models on the difficult task of predicting outcomes at the Supreme Court of the United States. It evaluates not only whether the models select the eventual winner, but also how consistently they perform across cases.
The results frame judicial forecasting as an evaluation problem rather than a demo: useful legal prediction requires transparent methods, careful validation, and a clear account of where model-generated reasoning remains unreliable.
Original publication
The full article lives at its original source. Use the outbound link below to open the complete publication.