OpenAI Claims Three Major AI Breakthroughs in Mathematics, Computer Science and Physics

OpenAI has announced three potentially significant scientific breakthroughs involving mathematics, computer science and theoretical physics, highlighting the growing ability of artificial intelligence to contribute to advanced research. Announced on October 26, 2026, the findings involve a mathematical conjecture connected to several major problems in harmonic analysis, a longstanding assumption about computational complexity, and the mathematical existence of certain quantum field theories. According to OpenAI, its internal AI models generated research results that could advance scientific understanding in areas where difficult questions have challenged researchers for years. The announcements represent an important development in the competition to build AI systems capable of producing original scientific work rather than simply summarising existing knowledge. However, the claims have not yet been independently established, and specialists must examine the underlying proofs before their scientific significance can be determined.

The first breakthrough concerns mathematics, where OpenAI says its models produced a proof of the local smoothing conjecture in every dimension of at least two. The company states that this result also has implications for the Kakeya conjecture and Stein’s restriction conjecture, important problems in harmonic analysis. These areas of mathematics investigate how functions, waves and frequency information behave, with connections to partial differential equations and other branches of mathematical research. OpenAI reports that its model generated the proof on October 18 and that the work was formalised the following day using Lean, a computer-assisted system for checking mathematical proofs. The company has released the mathematical write-up and formal proof for researchers to examine. If the argument withstands independent scrutiny, it could represent a substantial contribution to mathematical research and demonstrate how AI can assist with problems that require complex chains of reasoning.

The second claim involves computer science and the Exponential Time Hypothesis, commonly known as ETH. This longstanding conjecture concerns the computational difficulty of solving certain Boolean satisfiability problems and has been used to establish conditional lower bounds in numerous research papers. OpenAI says its AI system developed a deterministic algorithm that challenges the hypothesis by solving a relevant computational problem in subexponential time. The company reports that the model produced its proposed disproof on October 16 and that the result was formalised in Lean on October 17. Researchers Russell Impagliazzo, Ramamohan Paturi and Ryan Williams are expected to help organise an independent examination of the work. If the claim is correct, it could require computer scientists to reconsider some results that rely on ETH. Nevertheless, overturning this hypothesis would not automatically solve the P versus NP problem or mean that every computationally difficult task can suddenly be completed quickly.

The third announcement concerns theoretical physics, particularly quantum field theory, which provides mathematical foundations for describing fundamental particles and their interactions. OpenAI claims that its models produced rigorous existence proofs for certain four-dimensional quantum field theories, including quantum chromodynamics, the theory describing the strong interaction between quarks and gluons. The work also reportedly includes a proof concerning the existence of pure Yang-Mills theory, which is connected to an important unsolved problem recognised by the Clay Mathematics Institute’s Millennium Prize programme. Nobel Prize-winning physicist David Gross is expected to lead a group examining the physics results. Unlike the mathematics and computer science materials, the physics outputs have not been fully released publicly because the reviewing group has chosen to work on preparing the findings for scientific examination. The results therefore require particularly careful scrutiny before researchers can determine whether they resolve the relevant mathematical questions.

Ultimately, OpenAI’s three announcements highlight a potential shift in the role of artificial intelligence, from a tool that assists researchers to a system that may contribute directly to producing new scientific knowledge. The use of Lean provides an additional method of checking formal mathematical reasoning, but independent experts must still verify that the formal statements accurately represent the original problems and that the results are genuinely significant. Scientific acceptance depends on rigorous examination, reproducibility and scrutiny by specialists, not simply on a company’s announcement or a computer-generated proof. OpenAI says it is making the mathematical and computer science materials available for examination and supporting collaborations intended to clarify the findings. If the claims are validated, they could influence future research methods and encourage greater use of AI in advanced mathematics, computing and physics. Until that verification is complete, however, these developments should be regarded as promising claims rather than three definitively established scientific breakthroughs.