08-29-2026, 09:05 PM
Summary
The paper studies how artificial intelligence is being used in contemporary mathematical research by analyzing 32,944 arXiv papers submitted between March 1 and August 20, 2026. The authors identified 3,575 papers that explicitly disclosed AI use, with 1,712 involving substantive mathematical assistance such as proof construction, problem formulation, formalization, or verification. Substantive AI use rose sharply from $1.39%$ of mathematics submissions in March to $14.09%$ by August 20. Proof construction was the most common advanced use, appearing in 1,225 papers. Adoption varied strongly by field: Combinatorics had the largest number of AI-assisted papers, while Metric Geometry had the highest proportion. The study also found 717 records involving named open problems, of which 510 were described by the paper authors as completely resolved, either through proofs or counterexamples.
The study also shows that AI-assisted mathematics is concentrated geographically and technologically. The United States and China together accounted for about two-thirds of the recognized author-country contribution, while OpenAI systems were mentioned in about $60%$ of papers that disclosed AI use, followed by Anthropic, Google, and DeepSeek. The authors argue that mathematics is rapidly moving beyond using AI merely for writing, coding, or formatting toward using it directly in research and proof discovery. However, they stress that the reported solutions to open problems are claims made by the authors of those papers and are not independently verified, and that the study only captures researchers who explicitly disclose AI use. The results therefore provide an early but striking picture of how quickly AI is becoming integrated into mathematical research.
ARTICLE
The paper studies how artificial intelligence is being used in contemporary mathematical research by analyzing 32,944 arXiv papers submitted between March 1 and August 20, 2026. The authors identified 3,575 papers that explicitly disclosed AI use, with 1,712 involving substantive mathematical assistance such as proof construction, problem formulation, formalization, or verification. Substantive AI use rose sharply from $1.39%$ of mathematics submissions in March to $14.09%$ by August 20. Proof construction was the most common advanced use, appearing in 1,225 papers. Adoption varied strongly by field: Combinatorics had the largest number of AI-assisted papers, while Metric Geometry had the highest proportion. The study also found 717 records involving named open problems, of which 510 were described by the paper authors as completely resolved, either through proofs or counterexamples.
The study also shows that AI-assisted mathematics is concentrated geographically and technologically. The United States and China together accounted for about two-thirds of the recognized author-country contribution, while OpenAI systems were mentioned in about $60%$ of papers that disclosed AI use, followed by Anthropic, Google, and DeepSeek. The authors argue that mathematics is rapidly moving beyond using AI merely for writing, coding, or formatting toward using it directly in research and proof discovery. However, they stress that the reported solutions to open problems are claims made by the authors of those papers and are not independently verified, and that the study only captures researchers who explicitly disclose AI use. The results therefore provide an early but striking picture of how quickly AI is becoming integrated into mathematical research.
ARTICLE
┌────────────────────────────────┐
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘