OpenAI Says AI Produced Three Major Science Results
OpenAI says its internal models produced results spanning the Kakeya conjecture, a disproof of the Exponential Time Hypothesis and an existence proof for quantum chromodynamics, with independent experts now examining the claims.
OpenAI says its internal AI models have produced three new research results spanning mathematics, computer science and physics, including a claimed disproof of a hypothesis that computer scientists have believed for more than a quarter of a century. The company announced the results on October 26, 2026, less than a month after it released hundreds of manuscripts it said contained solutions to long-standing mathematics questions.
The claims are extraordinary, and OpenAI itself is framing them as work to be examined rather than settled fact. That distinction matters more than the headlines. Each of the three results now faces the same test: independent experts reading the proofs, checking them line by line, and deciding whether they hold up.
The three claims
The mathematics result concerns local smoothing for the Euclidean wave equation. OpenAI says the proof, which uses an auxiliary partial differential equation, establishes a result known as Sogge’s local smoothing for waves in every spatial dimension of at least two. From that, the company says, follow the Kakeya maximal and dimension conjectures and Stein’s spherical restriction estimates throughout the interior of the conjectured range. The Kakeya conjecture, about how thin tubes pointing in different directions can overlap, is one of the central open problems in harmonic analysis. OpenAI says it is releasing the model output directly, as it did in its October 6 release, for researchers to work through themselves.
The computer science result is the most startling of the three. OpenAI says its model produced a deterministic algorithm that exactly counts the satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space, and that the case of 3-SAT refutes the Exponential Time Hypothesis. The hypothesis, proposed in 1999, holds roughly that some hard computational problems cannot be solved much faster than brute force. If the disproof holds, it would overturn a foundational assumption of complexity theory. OpenAI has published the write-up and a Lean formalization on GitHub.
The physics result is a mathematically rigorous existence proof of certain interacting quantum field theories containing gauge fields and fermions in four dimensions, including quantum chromodynamics, the theory of the strong nuclear force, and a unified completion of the Standard Model of particle physics. Unlike the other two results, the physics material has not been released in full.
Shepherds, and a warning about shepherds
What is different this time, OpenAI says, is how the results are being handled. For the computer science and physics results, the company has enlisted external experts it calls shepherds. The computer science group is led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego, who were part of the group that originally proposed the Exponential Time Hypothesis in 1999, together with MIT professor Ryan Williams. At the group’s request, OpenAI made the materials public immediately so others can examine them, and the group has published an open letter asking the community for input on how the claim should be examined.
The physics group will be led by Nobel laureate David Gross, together with Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov. For the mathematics result, OpenAI says its advisers recommended releasing the AI output directly for researchers to explain themselves, rather than routing it through designated recipients first.
That shepherd model has already drawn a pointed objection. A panel of mathematics advisers OpenAI itself assembled, known as AGMAI, said the day before the announcement that it recommends AI companies do not adopt the practice of designating certain people to receive output first, citing unequal access, borrowed reputation and biased selection. Instead, the panel recommended that labs release output on scholarly repositories with Lean-formalized proofs, information on how problems were chosen, and funding for the people working to understand the results.
Why caution is the story
The history here counsels patience. AI-generated proofs of famous conjectures are only as good as their verification, and the verification of a result of this size is itself a research project. The computer science claim in particular would, if true, rewrite textbooks; that is precisely why it must survive scrutiny from the same community that built the hypothesis it claims to overturn, including two of the hypothesis’s own proposers.
OpenAI’s decision to publish the materials and invite examination is the right structure, and the involvement of Impagliazzo, Paturi and Williams gives the computer science claim a serious hearing. But no independent verification has been reported yet, and the physics result cannot even be fully examined until more of it is released. Until then, the accurate statement is not that AI solved three great problems. It is that OpenAI says its models produced three proofs, and that the scientific community is now checking them.
That is still a remarkable moment. A few years ago, the idea that an AI system would produce candidate proofs at this level would have seemed fanciful. That such proofs now exist in a form experts can test is itself a change in how science gets done. Whether they survive the test is the next chapter, and it belongs to the mathematicians, computer scientists and physicists reading them, not to the press release.