Epoch reports 25% of math preprints acknowledge AI use in August
Summary
The share of math preprints on arXiv acknowledging AI has surged from 4% in April to 25% in August 2026, reflecting a significant trend in the integration of AI within mathematical research. This increase persists even when considering papers with authors who have a consistent publication history prior to 2023. Notably, 6% of papers in August attributed substantial research ideas directly to AI, highlighting its evolving role beyond just tasks such as literature review and coding. The data is analyzed using a dual-model validation system for enhanced accuracy, ensuring that only reliable AI use cases are classified.
Analysis
Epoch: Epoch is an organization specializing in AI research, data analysis, and tracking technological trends in scientific fields. It created the AI Use in Math Research explorer to monitor acknowledgments of AI use across arXiv math preprints and classify different contribution types. This work directly produced the analysis showing growing AI involvement in recent mathematics research. arXiv: arXiv is a major open-access repository for scientific preprints, serving as a primary platform for sharing research in mathematics and related disciplines. The analysis examined every math preprint submitted to the platform over an extended period to identify patterns in AI acknowledgments. This dataset forms the foundation for insights into AI's expanding role in academic workflows. GPT-5.6 Sol: GPT-5.6 Sol is a large language model designed for advanced text classification and analysis tasks. It was applied to scan math preprints on arXiv and categorize acknowledgments of AI assistance across research and non-research use cases. The model supported the identification of AI contributions such as idea generation in the reported trends. Claude Fable 5: Claude Fable 5 is an AI system used for validation in automated text analysis pipelines. It reviewed classifications made by GPT-5.6 Sol regarding AI use acknowledgments in arXiv math papers to improve accuracy. The combination of models enabled reliable detection of both substantial research contributions and auxiliary tasks like coding or literature review. Research Trends: AI acknowledgments appear in a growing share of math preprints even after accounting for authors with long publication histories. AI Contributions: AI is credited with generating substantial research ideas in addition to supporting tasks such as literature review and coding. Analysis Methods: Dual-model validation with stricter checks on research-related categories improves the reliability of detected AI use cases.
Categories
aitechai_agentsmachine_learning