OpenAI announced results in mathematics, computer science and physics on Monday, and then did something the field had spent a month asking for: it slowed down and asked outside experts how to release them. The result has not calmed anyone.
The three claims are large. The mathematics output purports to resolve the Kakeya conjecture in all dimensions, the question of how little space is needed to turn a needle through every direction. A three-dimensional proof was among the work that earned NYU's Hong Wang a Fields Medal this year. A second result concerns quantum chromodynamics and Yang-Mills theory, the framework behind the strong nuclear force and one of the Millennium Prize problems, though OpenAI's work does not address the unresolved mass gap. The third claims to disprove the exponential time hypothesis, the assumption that certain problems cannot be solved much faster than brute force, a result adjacent to P versus NP.
How each was released differs, and that is the dispute. For mathematics, OpenAI published the AI-generated output and left the field to organise its own explanation. For computer science, it is assembling an external group to explain the result. The physics work has not been released at all; OpenAI says it will defer to outside experts on how to do that.
The company calls the advisers shepherds. Researchers call the arrangement gatekeeping. The Advisory Group on Mathematics and Artificial Intelligence, a body OpenAI had previously said it would consult, warned against the practice a day before the announcement: selecting who sees output first is unnecessary, borrows the credibility of the chosen, and risks unequal access, biased selection and delay. It recommended posting results to scholarly repositories with formal proofs and information about how problems were chosen, plus funding for people trying to understand them. After the physics result stayed closed, AGMAI urged OpenAI to publish the natural-language and Lean outputs without further delay, noting that by OpenAI's own account the formalisation is complete.
The individual reactions are sharper than the institutional ones. Constantin Kogler of the Institute for Advanced Study said it makes no sense that only a few distinguished scientists can look first. Andreas Karch, a physics professor at UT Austin, called the hand-picked group a disaster: the large labs have enough power already, and should not be picking winners and losers. He asked why the proof cannot simply be released now. Brendan Hassett of Brown said the Kakeya claim looks well supported but that he had not checked the Lean formalisation, and argued that many groups wrestling with results is exactly what speeds adoption of new tools.
Xiaoyu He of Georgia Tech put the deeper problem plainly: the results are impressive, but OpenAI is staging responsibility about release while the underlying issues stay fixed. Foundational results arrive faster than humans can contribute to them, in papers that are hard to read and hard to verify. That is the tension the shepherds cannot resolve. A disclosure process can be consultative or it can be open, and the researchers who have to check the work want both.
