ArXiv reports AI use in math papers surged to 14% in 6 months

by@rohanpaul_ai

Summary

Disclosed use of artificial intelligence in mathematics has surged from 1.39% to 14.09% in less than six months, according to an analysis of 32,944 math-related papers on arXiv published between March 1 and August 20, 2026. The significant increase in submissions that explicitly mention AI use is largely driven by proof construction, which is the most common category of AI contributions. Moreover, this rapid growth is unevenly distributed among subfields; Combinatorics has the highest number of AI-assisted papers, while Metric Geometry leads in the percentage of papers utilizing AI. Additionally, the research landscape shows a geographic concentration, with authors from the United States and China accounting for about two-thirds of the contributions.

Analysis

arXiv: arXiv is an open-access repository for electronic preprints across mathematics, physics, computer science, and related disciplines, enabling rapid dissemination of research ahead of formal publication. It serves as the primary platform for sharing and accessing new mathematical papers. The study examined thousands of mathematics submissions posted to arXiv to measure and analyze patterns of disclosed AI tool usage in research. Field Variation: AI-assisted mathematics research appears unevenly across subfields, with combinatorics accounting for the highest total volume of such papers and metric geometry showing the highest proportion within its category. Application Focus: Proof construction represents the most common category of substantive AI contributions in the analyzed mathematics papers. Geographic Concentration: AI-involved mathematics research shows strong geographic concentration, with the United States and China together accounting for the majority of weighted author contributions.

Categories

aimachine_learningtech
View Original Tweet