Researchers are getting tired of OpenAI gatekeeping its breakthroughs

OpenAI announced three results touching some of the hardest open problems in research. One claims to resolve the Kakeya conjecture in all dimensions. Another claims to disprove the exponential time hypothesis in theoretical computer science. A third, in physics, concerns quantum chromodynamics and Yang-Mills theory, and has not been released at all.

Experts told The Verge the work deserves to be taken seriously. But the release plan, guided by handpicked expert "shepherds," gives different fields different levels of access. That is drawing sharp criticism from researchers who say announced results should be open to everyone, right away.

Let's start with the good news, because it is genuinely big. If these results hold, AI systems are now contributing to problems that have stumped mathematicians and scientists for decades. Brown's Brendan Hassett said the math claim looks well supported, while noting he had not yet checked the formal verification. That is the kind of careful optimism the moment calls for. We are pro this kind of AI. Tools that push the frontier of knowledge are exactly what we hope the technology delivers.

But science runs on scrutiny, and scrutiny requires access. Announcing a result while restricting the materials to experts the company picks creates two problems. It slows verification, and it lets a powerful lab decide who gets first credit and influence in a field. UT Austin physicist Andreas Karch put it bluntly, saying the big labs already have enough power and should not be picking winners and losers. The Advisory Group on Mathematics and AI, which OpenAI had pledged to consult, had warned against exactly this practice a day earlier, and then urged the company to release the physics outputs now that its formal verification is complete.

To be fair, OpenAI's previous approach drew criticism too. Earlier this month, researchers were swamped by hundreds of preprints at once. A smaller batch with more thought behind it is an improvement, and some researchers said so. Stanford's Jared Duker Lichtman made the reasonable point that any approach will upset someone.

The deeper issue, raised by Georgia Tech's Xiaoyu He, is that no release strategy fixes the core tension: fundamental results are arriving fast, without much human contribution, in papers that are hard to read and verify. That is where this leads if it scales. Fields could end up dependent on a few labs to generate results and on those same labs to decide how they are explained.

Our read: release the materials publicly, keep the expert groups to help explain them, and fund the people doing the verification work. Openness is how these results become trusted knowledge rather than corporate announcements.

A proof of the three-dimensional Kakeya conjecture was among the work that earned NYU mathematician Hong Wang a Fields Medal this year. The exponential time hypothesis is closely tied to P versus NP, and Yang-Mills theory sits at the center of another Millennium Prize problem, though OpenAI's work does not address the unresolved mass gap.

Lean is a system for formally verifying proofs by computer, and it has become central to how AI-generated math is checked.