09-21-2026, 07:48 PM
The paper The Gold Rush in AI4Math: Where Are We Now? examines how artificial intelligence is actually being used in mathematical research, rather than merely testing what AI could theoretically do. The authors analyzed 32,944 mathematics submissions to arXiv from March 1 to August 20, 2026, identifying 3,575 papers that explicitly disclosed AI use, of which 1,712 involved substantive mathematical contributions.
The proportion of mathematics papers with substantive AI use rose sharply from 1.39% in March to 14.09% by August 20. AI involvement is highly uneven across fields: Combinatorics has the largest number of AI-assisted papers, while Metric Geometry has the highest proportion. Proof construction is the most common substantive use, followed by formalization/verification and other research assistance.
A particularly striking finding is that AI is already being used on open mathematical problems, not simply routine exercises or editing. The study identifies 717 named open-problem records associated with substantive AI use; 510 are described by their authors as fully resolved, although the paper emphasizes that these resolutions have not been independently mathematically verified by the study. The research also finds strong geographic and technological concentration, with the United States and China accounting for roughly two-thirds of weighted country representation, while OpenAI systems are reported most frequently,
followed by Anthropic. Overall, the authors conclude that AI-assisted mathematics is expanding extremely rapidly but remains at an early, heterogeneous stage, and they stress that their methodology—relying on authors' explicit disclosures—likely underestimates actual AI use because undisclosed usage cannot be observed.
ARTICLE
The proportion of mathematics papers with substantive AI use rose sharply from 1.39% in March to 14.09% by August 20. AI involvement is highly uneven across fields: Combinatorics has the largest number of AI-assisted papers, while Metric Geometry has the highest proportion. Proof construction is the most common substantive use, followed by formalization/verification and other research assistance.
A particularly striking finding is that AI is already being used on open mathematical problems, not simply routine exercises or editing. The study identifies 717 named open-problem records associated with substantive AI use; 510 are described by their authors as fully resolved, although the paper emphasizes that these resolutions have not been independently mathematically verified by the study. The research also finds strong geographic and technological concentration, with the United States and China accounting for roughly two-thirds of weighted country representation, while OpenAI systems are reported most frequently,
followed by Anthropic. Overall, the authors conclude that AI-assisted mathematics is expanding extremely rapidly but remains at an early, heterogeneous stage, and they stress that their methodology—relying on authors' explicit disclosures—likely underestimates actual AI use because undisclosed usage cannot be observed.
ARTICLE
┌────────────────────────────────┐
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘