8,000 Mathematicians Object to How AI Labs Announce Proofs

A declaration accusing AI companies of damaging mathematics had collected 8,047 public endorsements when its list was checked on 28 September 2026. The text, 'A Severe Misalignment of AI in Mathematics', was published on 11 September with a digital object identifier and 28 Fields Medallists among its initial signatories, including Terence Tao, Peter Scholze, Maxim Kontsevich, Pierre Deligne and Cedric Villani.

The complaint is not that models cannot do mathematics. The signatories accept that large language models now solve major open problems. Their objection is to what the pursuit of those problems as benchmarks does to the discipline that has to absorb the results.

What OpenAI says it is holding back

OpenAI supplied the immediate provocation. In a post dated 21 September 2026, the company said it had begun training a new internal model on 28 August and that the model had since resolved more than 100 long-standing open problems across most areas of mathematics. The post states that the pace surprised the mathematicians inside OpenAI. None of those results has been published, and the model is not available to the public.

That followed a larger claim. Quanta Magazine reported on 8 September 2026 that OpenAI mathematicians, directing some 10,000 autonomous agents on a non-public model, had found a singularity in the three-dimensional Navier-Stokes equations, resolving one of the six remaining Millennium Prize Problems set by the Clay Mathematics Institute in 2000. That result, unlike the 100 that followed, was formally verified in the Lean proof language.

The attribution problem has a name

The charge about citing previous work has a concrete illustration in the Navier-Stokes case. Quanta reported that both AI-assisted efforts, OpenAI's and a separate one by Tristan Buckmaster of New York University with Levent Alpoge of Anthropic, leaned heavily on a strategy developed by Diego Cordoba of the Institute for Mathematical Sciences in Madrid and Luis Martinez-Zoroa of CUNEF University. Charles Fefferman of Princeton University, who wrote the Clay Institute's official description of the problem, named those two as the heroes of the story. Buckmaster wrote in his own announcement that he believed Martinez-Zoroa deserved a Fields Medal.

A group that cannot ask for a pause

OpenAI's response was to point at an advisory group of nine mathematicians hosted at the Institute for Advanced Study in Princeton: Francois Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten and Melanie Matchett Wood. The group takes no payment, may publish advice OpenAI did not request, and may comment on the company's effect on mathematics. One boundary is drawn in OpenAI's own words: the group will not be responsible for advising it on how to pace its internal progress in mathematics.

The group's own site describes its origin differently from OpenAI's framing. It says it came together after OpenAI approached some members about establishing an external advisory board, and that they decided instead, in agreement with OpenAI, to create an independent group. Its stated task is narrow: advising OpenAI on how to coordinate the release of a large number of significant results that the company reports its internal model has produced.

One name sits on both documents. Martin Hairer, who won the Fields Medal in 2014, signed the declaration and joined the group now advising OpenAI. That overlap is the most useful signal in the episode, because it suggests the working assumption among senior mathematicians is that the flow of results cannot be stopped and the only live question is sequencing.

What the declaration does not do is set a standard anyone can be measured against. It asks for proper writeups, attribution and time, without specifying how long, in what venue, or who adjudicates a dispute, and it has no mechanism beyond signatures. Against that, the Navier-Stokes proof was machine-checked in Lean before publication, which is a stricter verification than most human papers receive at announcement, so the community's objection is about scholarly process rather than about correctness.