Fields Medal Winner Says He Cried After OpenAI Released 700 AI Math Papers: ’They Have Annihilated My Field’

Hugo Duminil-Copin, the French mathematician who won the 2022 Fields Medal, says he cried after OpenAI released hundreds of AI-generated mathematical manuscripts on 6 October 2026, describing the development as a threat to his research field.

Speaking during a video conference organised by the Société Mathématique de France on 8 October, he warned that the rapid arrival of machine-generated research could fundamentally change how mathematicians work and share ideas.

OpenAI's release initially contained 722 manuscripts grouped into 372 families, with three subsequently withdrawn, leaving 719.

The company presented the work as addressing numerous open mathematical problems, but the results have not all been independently verified.

For researchers, the controversy is about more than whether AI can produce a proof. It also concerns who gets credit, how results are checked and what happens to the human collaboration that has long driven mathematical discovery.

OpenAI's AI Math Papers Leave a Fields Medal Winner Distressed

Duminil-Copin described the personal impact of the announcement during his presentation. 'Tuesday night, I cried and it woke up my daughter,' he said, according to Le Monde's reporting.

He later shared a slide depicting a rabbit being crushed by several trucks, an image generated by AI. The illustration captured his sense of being overwhelmed by the speed and scale of the release.

'They have annihilated my field of research,' he said. He also accused the company of wanting to demonstrate that it could 'massacre us', although that was his interpretation of the release rather than established evidence of OpenAI's intentions.

The announcement followed weeks of mounting concern among mathematicians.

In September, OpenAI said its systems had produced a proposed solution to the long-standing Navier-Stokes problem, which has carried a million-dollar prize since 2000.

That claim was another shock for Duminil-Copin, who had already encountered evidence that AI systems could tackle difficult problems in his area of research.

Anthropic, an OpenAI competitor, had also presented a solution to a prominent problem in his field. Duminil-Copin and colleagues were themselves close to solving it, but chose to examine the work carefully rather than immediately accept the result.

The October release intensified those concerns. In one research programme connected to his work, OpenAI's agents produced 13 papers covering successive stages of a larger project.

The scale of the output challenged the assumption that these systems were merely rearranging familiar ideas rather than making contributions to research.

Yet the number of manuscripts should not be confused with the number of conclusively established breakthroughs.

OpenAI's repository acknowledges that the results are at different stages of verification and that some proofs may contain problems. Three manuscripts were withdrawn after errors were identified, while other papers were revised.

AI Math Papers Raise Questions About Verification and Credit

The way the research was published has become almost as contentious as the mathematical claims themselves.

On 21 September, OpenAI said it had heard criticism from mathematicians, including Fields Medal winners, and promised to work with an international advisory group hosted at the Institute for Advanced Study in Princeton.

The group issued recommendations on 29 September, but the 6 October release prompted renewed questions about whether those recommendations had been followed.

The advisory group did not endorse the release process. It stressed that the mathematical community would need to assess the results and determine how far its recommendations had been implemented.

Concerns included the use of a proprietary internal AI model and the limited application of formal verification.

Around 42 per cent of the top-line results had Lean formalizations, according to the repository's overview.

Lean is a system used to express mathematical proofs in a form that can be checked by a computer. Its use can provide an important safeguard, although formalization is not the only way to assess whether a mathematical argument is sound.

Some researchers also found the manuscripts difficult to read, even for specialists. The sheer volume presents its own problem.

A result may be promising, but mathematicians still need time to understand its reasoning, check its validity and establish how it fits into existing work.

Duminil-Copin said he was not eager to read the more than 3,500 pages generated by AI that relate to his field. The experience, he explained, lacked the pleasure of reading work produced by human colleagues whose faces he knew.

'OpenAI does not care about the human adventure,' he said. 'What has been done is even inhuman.'

Young Mathematicians Face an Uncertain Future

In a telephone interview the following day, Duminil-Copin said his distress was not only about his own career or that of his colleagues.

He was particularly concerned about younger researchers, who could find it harder to establish themselves if AI systems rapidly produce results that would once have taken years of human investigation.

Mathematics depends on more than the final proof.

Researchers exchange tentative ideas, discuss unfinished problems and learn from experienced colleagues. Duminil-Copin fears that mathematicians may become reluctant to share promising questions publicly if companies can use those discussions to direct their AI systems towards the same problems.

He warned that the culture of open exchange was being shattered. Some mathematicians, he said, had already begun considering restrictions on what they share.

Despite his concerns, Duminil-Copin intends to examine the AI-generated work, continue pursuing mathematical problems without AI and devote more time to teaching and public outreach.

He wants researchers to find ways to understand and communicate machine-generated results collectively, rather than allowing the papers to remain a flood of largely unread material.

The challenge is no longer simply whether AI can produce mathematical work. It is whether researchers can verify and absorb that work without losing the open exchange of ideas that helped make such discoveries possible.