epoch.ai web signal

Epoch AI tracker: math preprint AI acknowledgments hit 25%

TL;DR

  • Acknowledgments of AI use in arXiv math preprints rose from 4% in April 2026 to 25% in August, per Epoch AI's tracker.
  • About 6% of August papers credit AI with a substantial research contribution, Epoch's top-tier category.
  • Differential geometry led the surge, climbing from roughly 8% to 57% in four months.

Acknowledgments of AI use in math preprints submitted to arXiv rose from 4% in April 2026 to 25% in August, according to a tracker from Epoch AI. About 6% of August papers credit AI with a "substantial research contribution," which Epoch sorts above softer categories like research assistance, writing, code and computation, formalization, and literature review.

The jump is uneven. Differential geometry climbed from roughly 8% to 57% of papers acknowledging AI use over the same four months.

Across the 56.2k math papers in the full dataset, combinatorics (6.6k) and analysis of PDEs (6.4k) are the biggest sub-corpora. Epoch pipes each flagged preprint through an LLM-based classifier with a second model acting as validator; a human review of 250 use-case classifications "agreed with 93% of classifications in total." The underlying CSV, last updated October 6, 2026, is released under a Creative Commons Attribution license.

Shared on Bluesky by 2 AI experts