OpenAI claims three new AI breakthroughs in math, physics and computer science

OpenAI has published AI-generated work on mathematical conjectures, quantum field theory and a claim that challenges a long-standing assumption in computer science. The releases come with a new approach to expert scrutiny, even as the company faces questions about selective access, academic credit and how researchers should assess AI-generated discoveries.

OpenAI has unveiled three further research results generated by its AI models, including a claim that would overturn a widely accepted theory about the computational difficulty of solving certain problems. The work spans mathematical analysis, computer science and theoretical physics, but the company is taking different approaches to making the findings available for scrutiny.

The latest disclosures follow an October 6 release in which OpenAI published hundreds of AI-generated mathematical manuscripts for researchers to examine. This time, it has brought in established academics to help assess and explain two of the results, while leaving the mathematical work open for independent researchers to interpret.

The announcements raise a broader question for the scientific community -- how should researchers handle discoveries produced by AI systems when verifying the underlying reasoning, explaining the results and assigning credit can require substantial work of their own?

AI challenges established theories in maths and computer science

The mathematical result concerns local smoothing for the Euclidean wave equation, a problem in the analysis of wave propagation. OpenAI says its model produced a proof covering every spatial dimension of two or more. According to the company, the result also establishes the Kakeya maximal and dimension conjectures, alongside Stein's spherical restriction estimates across the interior of their predicted range.

These connections could make the work significant beyond the original problem. However, the release consists of the model's output rather than a fully explained, independently validated research paper. OpenAI has invited mathematicians to examine the material and develop their own accounts of the reasoning.

The Kakeya conjecture has attracted particular attention in mathematical research. NYU professor Hong Wang received a Fields Medal in 2026 for work resolving the conjecture in three dimensions, according to the background provided by OpenAI's announcement. The company's broader claim covers results in every dimension, although the new output still requires careful examination by specialists.

The computer science result makes a more direct challenge to an established assumption. OpenAI says its model has found a deterministic algorithm that counts satisfying assignments for fixed-width conjunctive normal form (CNF) formulae in subexponential time and polynomial space.

For the three-variable satisfiability problem, commonly called 3-SAT, the result would disprove the Exponential Time Hypothesis (ETH). The hypothesis, proposed by Russell Impagliazzo and Ramamohan Paturi in 1999, holds that 3-SAT cannot be solved in subexponential time.

ETH has played an important role in theoretical computer science, helping researchers establish limits on how efficiently computational problems can be solved. If OpenAI's claim withstands scrutiny, it could require researchers to reconsider results that rely on the hypothesis. For now, the announcement should be treated as a claim awaiting verification, rather than a settled change to computer science.

Experts to examine results as debate over access grows

OpenAI has arranged for a group led by Impagliazzo, Paturi and MIT professor Ryan Williams to investigate the computer science proof. The company says the researchers will help verify and explain the result. At the group's request, OpenAI has made the AI-generated write-up and a Lean proof available on GitHub, alongside an open letter inviting feedback on how the claim should be assessed.

The physics work is being handled differently. OpenAI describes it as a mathematically rigorous existence proof for certain interacting quantum field theories in four dimensions, involving gauge fields and fermions. The theories include quantum chromodynamics (QCD), which describes the strong interaction, and a unified completion of the Standard Model of particle physics.

The full physics material has not yet been published. OpenAI says David Gross will lead a group that includes Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov in examining the work and helping explain its implications.

The decision to involve selected experts has itself become a point of contention. An advisory panel convened by OpenAI, known as AGMAI, recently recommended that AI companies avoid giving designated individuals privileged early access to generated research. Its concerns include unequal opportunities for researchers, the use of established academics' reputations to lend credibility to unverified claims, and the risk of bias in choosing who gets involved.

Instead, the panel recommended that AI laboratories publish their results through scholarly repositories, provide Lean-formalised proofs where possible, explain how they selected the problems and fund researchers who undertake the work of understanding the output.

OpenAI's accompanying letter asks the research community to help establish standards for disclosure, conflicts of interest and the use of unpublished material. It also raises the question of how academic recognition should account for the people who verify, correct, simplify and explain AI-generated findings, rather than only the system that first produces them.