OpenAI’s feud with mathematicians is only escalating

An open letter from twenty-five Fields Medal winners argues that AI labs are treating landmark math problems as scoreboard trophies—announcing solutions in a rush, short-circuiting the writeups, method isolation, and citation that turn a result into living mathematics. TechCrunch reports the letter arrives amid fresh friction: NYU’s Tristan Buckmaster accused OpenAI of pressuring him not to credit an Anthropic collaborator on an important problem, and of possibly using collaborators’ Codex work en route to its own marathon-weekend proof claim. That proof, the piece notes, remains unverified.

On Thursday, OpenAI withdrew sponsorship of a Caltech math event after criticism from researchers there. Separately, mathematicians are growing paranoid that their own Codex use may feed future OpenAI models—while labs with tens of millions of dollars in inference spend can try to beat original researchers to a proof. The Leiden Declaration in June already sketched recommendations for mathematicians, institutions, and policymakers. This letter sharpens the cultural fight: who owns the credit, and who keeps the human transmission chain alive.

We’re not anti-AI-for-math. A model that helps crack hard problems can be a genuine gift—if the proof can be checked, taught, and woven into the canon. The Fields Medalists’ warning is sharper than “AI stole my thunder.” It’s that rushed announcements without writeups and attribution turn creative work into a black box, raise plagiarism questions, and leave human mathematicians as unpaid janitors for AI-conceived ideas that never fully become “alive.”

The Buckmaster episode is the interpersonal flashpoint. If labs lean on researchers’ tooling and then squeeze credit away from rival-affiliated collaborators, open collaboration dies first. OpenAI’s Caltech sponsorship pullback after criticism reads like reputation triage, not reconciliation. Useful tooling and opaque racing culture are not the same product.

The deeper trajectory: if frontier labs can outspend original researchers on inference whenever a useful path appears, secrecy becomes rational. Students lose the apprenticeship path that makes the next generation of mathematicians. Other fields—software, science, the creative trades—get a preview. The letter’s own closer is the point: the issue isn’t only math; it’s whether we keep sight of what the work was meant to achieve once AI changes how the work gets done.

I’m watching whether OpenAI’s claimed proof gets a proper, citable writeup the community can verify; whether labs publish attribution norms that survive competition with Anthropic and peers; and whether institutions start requiring human-readable proofs and clear credit trails before crowning an “AI solution.” Pro-useful-AI still means defending the culture that makes those solutions matter.

TechCrunch’s September 11 piece by Tim Fernholz covers the Fields Medalist letter, Buckmaster’s accusations toward OpenAI, the Caltech sponsorship withdrawal, and the earlier Leiden Declaration from a June working group of mathematicians. The letter argues AI solutions only help humanity if they can be understood and communicated by mathematicians and, ultimately, everyone else.