Glossary
October 26, 2026

OpenAI Says Its AI Has Overturned a Long-Held Belief in Computer Science and Solved a Major Math Problem, and Outside Experts Are Now Checking

OpenAI says an AI program it hasn't released has produced proofs that would settle two famous open questions, one in computer science and one in mathematics, plus a third in physics that it is holding back for now. If they hold up, they are among the biggest results AI has claimed so far. Independent researchers have only just started checking them.

OpenAI today posted two new mathematical papers on GitHub, the public website where it has been sharing work written by an AI program it uses internally and has not released. The company says each one settles a question researchers have worked on for decades.

The first concerns a widely held belief in computer science called the Exponential Time Hypothesis. Roughly, it says that certain logic puzzles take a hopelessly long time to solve as they get bigger, no matter how clever the method. Nobody has ever proved it, but researchers have treated it as true and built on it in hundreds of papers, in fields from biology to databases. OpenAI says its program found a method that solves these puzzles faster than the belief allows, which would mean the belief is false. The company says the program produced the argument on October 16. Even if it is right, it would not by itself make everyday software faster; working out what it means in practice will take more research.

The second is in pure mathematics. OpenAI says its program proved a statement about how waves spread out, first proposed by the mathematician Christopher Sogge in 1991, in every number of dimensions. If correct, that would also settle two related questions, including one called the Kakeya conjecture, which mathematicians only managed to prove for three dimensions last year. OpenAI says the program produced this proof on October 18.

OpenAI says both proofs were checked by Lean, a program that verifies every logical step by computer, within a day of being written. That is a strong check, but a proof that passes Lean is only as good as the way the question was typed into it, and outside experts have not yet gone through the work. The company has handed the computer science result to an independent group led by Russell Impagliazzo and Ramamohan Paturi of the University of California, San Diego, and Ryan Williams of MIT. At the group's request, the AI's materials were made public right away, and OpenAI has agreed to pay for a conference on the claim. OpenAI also says it has a third result, in physics, about the forces that hold the centers of atoms together, including part of one of mathematics' million-dollar Millennium Prize problems. A group of physicists led by David Gross is studying it and will explain it in their own words before the AI's work is released.

There are reasons for caution. These are claims, not yet confirmed results. Earlier this month, OpenAI withdrew three of the AI-written papers it had posted after finding a sign error that broke their arguments, and revised fourteen others. OpenAI itself calls the AI's write-ups "raw research data" that still has to be turned into human understanding, and it says it doesn't yet know the best way to release results like these.

Sources

  1. OpenAI — openai/math update log, October 26 release (GitHub)github.com · primary
  2. OpenAI — Lean formalization notes for "The Exponential Time Hypothesis is False" (GitHub)github.com · primary
  3. Interesting Engineering — OpenAI claims major proofs in math, computing and physicsinterestingengineering.com
  4. Unite.AI — OpenAI says its models produced three proofs in math, CS, and physicsunite.ai
Follow the timeline
Get an email when new entries are added.
© 2026 Sugarpine