OpenAI presents three scientific findings from its AI

OpenAI has announced three results generated by its models in mathematics, theoretical computer science and physics. The claims include a proof of the Kakeya conjecture in every dimension, a refutation of the Exponential Time Hypothesis and a mathematical proof concerning the existence of quantum theories such as quantum chromodynamics.

The company does not present these works as final conclusions for the general public. It describes them as raw research data that need to be reviewed, explained and put into context by external specialists.

Three results, three ways to publish them

OpenAI consulted researchers to decide how to share each finding. The result is an experiment with three different collaboration models:

• In mathematics, it is publishing the text generated by the AI and its formalization directly.
• In theoretical computer science, it is forming an external group to examine and explain the result while releasing all the materials.
• In physics, it is allowing a group of experts to decide which materials to publish and when.

The formalization is carried out in Lean, a system that allows a mathematical proof to be expressed in a language that can be checked automatically. This does not replace scientific review, but it helps detect invalid steps or formal errors.

OpenAI says the collaborators will not sign nondisclosure agreements, will be able to publish their conclusions and will not be paid for participating. They also will not be credited with the original discovery. Their recognition will relate to the work of understanding, explaining and developing new consequences from it.

A new proof about waves and geometry

The mathematical result addresses the Sogge local smoothing conjecture, formulated in 1991. In simple terms, it studies how waves concentrate and holds that the most extreme concentrations must last for only a short time.

The conjecture has important consequences in other areas of mathematical analysis. These include the Kakeya conjecture and the Stein restriction conjecture, two problems related to the geometry of sets and the behavior of waves.

OpenAI says its models produced a proof for all dimensions. According to the company, this would also resolve those two conjectures in every dimension. The result was generated on October 18, 2026, formalized in Lean the following day and is already available together with the explanatory text produced by the AI.

AI challenges a central computer science conjecture

The second result concerns the Exponential Time Hypothesis, a conjecture widely used in computational complexity. This branch of computer science studies how many resources an algorithm needs to solve a problem.

The hypothesis holds, in simplified terms, that certain problems cannot be solved much faster than the best-known methods allow. For years, it has been used to prove conditional limits in fields such as graph theory, computational biology and databases.

OpenAI says its system found a refutation of the hypothesis and that the result was later formalized in Lean. The natural-language text was generated on October 16, 2026, and the formalization was completed on October 17.

The finding will be reviewed by a group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams, specialists in algorithms and complexity theory. If the refutation withstands analysis, it could reopen the search for faster algorithms. That does not mean practical versions of those algorithms exist yet.

A possible advance in mathematical physics

The third work deals with quantum field theory, the mathematical framework used to describe particles and fundamental forces. OpenAI says its models produced a rigorous proof of the existence of certain theories with gauge fields and fermions in four dimensions.

The result would include quantum chromodynamics, which describes the strong force between particles such as quarks and gluons, as well as a possible unified extension of the Standard Model. It also includes a result concerning the existence of pure Yang-Mills theory, which is related to one of the Millennium Problems of the Clay Mathematics Institute.

The work will be supervised by a group led by David Gross, together with Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov. Unlike the mathematical result, OpenAI says it will not publish the materials produced by the AI for now. The group will first work on its own exposition.

The Yang-Mills proof was formalized on October 16, 2026, after the models generated results on Yang-Mills theory and quantum chromodynamics in September.

What changes for you

These announcements will not turn into an app on your phone tomorrow. Their importance lies elsewhere: if the proofs are correct, they could change basic knowledge about algorithms, geometry and particle physics.

The main challenge is independent verification. Having an AI generate a proof and seeing it pass through Lean provides a meaningful check, but other experts still need to understand the reasoning, review its assumptions and determine its consequences.

OpenAI will also fund conferences and workshops through a contribution of $5 million to the Simons Foundation. The company presents these groups as a way to accelerate research without claiming academic credit for the discoveries.

What matters now is not only whether the proofs are correct, but also how the scientific community responds: who can examine them, how long it takes to understand them and whether they lead to useful new ideas. AI can already propose results that appear to sit at the frontier of knowledge. The next step is to show that those results survive human scrutiny.