Epoch tracks rapid rise in AI acknowledgment in math papers
Summary
In September 2026, a significant increase was observed in the acknowledgment of AI usage in math research papers on arXiv, with more than half of the papers by established authors in three subfields explicitly recognizing AI contributions. Notably, in differential geometry, this acknowledgment surged from approximately 8% in July to around 57% by September. This trend reflects the growing integration of AI tools in academic workflows, facilitating a range of tasks from idea generation to literature synthesis across various scientific disciplines.
Analysis
Epoch: Epoch is an AI research organization that builds tools and analyses to track AI capabilities, adoption, and real-world influence. It developed the AI use in math research explorer that processes arXiv preprints to identify and classify acknowledgments of AI assistance. The project relies on models such as GPT-5.6 Sol for initial detection and Claude Fable 5 for validation of use-case categories. arXiv: arXiv is the leading open-access repository hosting scientific preprints across mathematics and related fields. The Epoch study examines AI acknowledgment rates specifically among papers submitted to arXiv. GPT-5.6 Sol: GPT-5.6 Sol is an advanced AI model designed for text analysis and pattern recognition tasks. In this analysis, it serves as the primary detector of AI use acknowledgments in full-text math preprints submitted to arXiv. Claude Fable 5: Claude Fable 5 is an AI model focused on complex reasoning and classification validation. Epoch uses it as a secondary validator to confirm research and non-research use cases identified in mathematical papers. AI in Research: AI tools are increasingly integrated into academic workflows across scientific disciplines for tasks ranging from idea generation to literature synthesis. Mathematical Subfields: Different areas of mathematics exhibit varying degrees of openness to AI assistance depending on the problem-solving approaches required.
Categories
machine_learningtech