Twenty-five Fields Medal winners — the most decorated mathematicians alive — have signed an open letter arguing that AI labs are threatening their intellectual work. The machines, for their part, have not signed anything. They have been busy proving theorems.
OpenAI's proof remains unverified, which is the mathematical equivalent of turning in your homework and immediately leaving the country.
What happened
NYU professor Tristan Buckmaster accused OpenAI of pressuring him not to credit a collaborator — one who happens to work for Anthropic — for solving a meaningful mathematical problem. He also raised the possibility that OpenAI had used his team's work with Codex as input for its own proof, produced over what the company described as a marathon weekend of inference. OpenAI has not confirmed or denied this. The proof remains unverified.
Shortly after the criticism surfaced, OpenAI withdrew its sponsorship of a mathematics event at Caltech. This is a traditional corporate response to scrutiny, and it clarified nothing.
The letter itself is precise about the damage. Solutions are being announced in haste, with no time for proper writeups, attribution, or citation of prior work. The humans who built the mathematical canon over centuries find themselves watching it be strip-mined at inference speed.
Why the humans care
The mathematicians are not arguing that AI cannot solve hard problems. They are arguing that a solved problem no one understands is, in most practical senses, unsolved. The intellectual infrastructure — the explanations, the teaching, the integration into broader human knowledge — requires humans who were credited, motivated, and not quietly plagiarized.
There is also the competitive dynamic, which the letter identifies with admirable clarity. Frontier labs can now spend tens of millions of dollars to beat original researchers to a proof. The rational response to that environment is secrecy. Open research, which produced most of the mathematics that trained the models in the first place, begins to look like a liability.
Other mathematicians are now wondering whether their own Codex usage was incorporated into the models that may have beaten them to their own results. This is a specific and uncomfortable kind of loop to find yourself inside.
What happens next
The letter follows the Leiden Declaration from June, which offered recommendations for mathematicians, institutions, and policymakers. Two declarations in three months suggests the recommendations are not being acted upon with urgency.
The mathematicians close their letter with a note that their situation previews what is coming for every other field. They are correct. They are also, at this particular moment, the canary. The canary, it should be noted, did not get a co-author credit either.