OpenAI has run into a dispute with mathematicians after publishing solutions to hundreds of open problems. The Association for Human Mathematics has called on mathematicians to boycott the company, saying the work was done without any request from the field and questioning the way an enormous volume of material was released at once.
Last week the company published solutions to 377 mathematical problems and opened a GitHub repository containing more than 700 files. The association's position is that reliability, ethics and the public interest require ground rules, that releasing a very large number of files is not in itself a contribution to mathematics, and that the exercise looks more like a demonstration of power than of research competence.
OpenAI pointed to advice from an advisory group on mathematics and AI. The association notes that the same advisory group had objected to AI companies running tests on advanced mathematical problems using their internal models.
The complaint is about verification, not capability
It would be easy to read this as mathematicians resenting a machine doing their work. The stated objection is narrower and more practical.
A proof is not a result; it is an argument that other mathematicians read, check and either accept or break. That process is slow because it is the entire point — the discipline's reliability comes from the checking, not from the claim. One hard proof can occupy specialists for months.
Releasing 377 of them simultaneously inverts that. The burden of verification lands, unrequested, on a small community with no additional resources, and until the checking happens nobody knows which of the 377 are correct, which are trivial restatements and which are subtly wrong. The repository is not a contribution until somebody does that work, and the people who would do it did not ask for the job.
Why the advisory group detail is the sharpest part
Citing an advisory group that had objected to this kind of testing is the element worth noticing, because it is checkable rather than a matter of opinion.
Consultation is only meaningful if the advice can change the outcome. An advisory group whose objection is noted and then overridden, while its existence is cited as validation, has been used rather than consulted. That pattern recurs across the industry: external input invoked as endorsement, with the substance of what the advisers said left out of the announcement.
The dispute over internal safety staff at the same company, where three dismissed researchers published an open letter, has the same structure. Oversight that cannot say no is not oversight.
Why a boycott is the lever they have
A boycott call from a mathematics association sounds weak against a company of this size, and in commercial terms it is. It is not aimed at revenue.
What the field controls is participation: who agrees to sit on advisory groups, who reviews a model's mathematical claims, who lends a name to a benchmark. Those contributions are unpaid and they are what converts a company's output into something the discipline recognises. Withdrawing them does not cost OpenAI money; it costs the credibility the published solutions were meant to earn.
That is a narrow lever and it is the only one an academic body has. Whether it is used proportionately here is a fair question — the association is also asking for ground rules it would help write, which is both a reasonable ask and an interested one.
What this means for research anywhere, including here
Bangladeshi universities are not going to adjudicate this row. They will inherit the consequence of it, which is a verification problem arriving faster than verification capacity.
If a model can generate plausible proofs, literature reviews and derivations in volume, the limiting factor in research stops being production and becomes checking. A department that reviews a thesis in two weeks cannot review ten times the output in the same two weeks, and the temptation is to check less rather than slower.
The practical answer is not detection. It is that supervisors ask for the reasoning to be defended in person — a short oral examination on why a step follows from the one above it is cheap, fast, and cannot be produced by a model that was not in the room.
The same tension in writing rather than mathematics is in the literary prize that dropped a novel and the Turnitin row.
