Did OpenAI's AI really disprove the Exponential Time Hypothesis?

OpenAI says so, and says the proof was verified in Lean, but independent review is still under way and the reviewers say their involvement is not certification.

What is the Kakeya conjecture?

A long-standing problem in harmonic analysis about how thin sets can point in many directions; OpenAI says its local smoothing proof resolves it in all dimensions.

Did OpenAI solve a Millennium Prize problem?

Only partly: it claims pure Yang-Mills existence, one half of the Yang-Mills Existence and Mass Gap problem; the mass gap is not claimed and the physics paper is not yet public.

Can I use any of this in my business today?

No. Nothing announced is a product, and no tool, price or release date was given.

When will we know if the claims hold up?

When the independent shepherd groups publish their verdicts; there is no date yet, so treat the claims as unverified until then.

OpenAI AI Science Discoveries: 3 Big Claims Explained

OpenAI Says Its AI Cracked Three Science Problems — Verification Still Pending

OpenAI says its internal AI models have produced three new scientific results — one each in mathematics, computer science and physics — including a claimed disproof of a hypothesis that hundreds of research papers rely on. The company announced the results on October 26, 2026, and the story has dominated AI coverage since. But there is a catch every reader should hold onto: none of the three claims has yet passed independent peer review.

What happened

In mathematics, OpenAI says its system proved the local smoothing conjecture in every dimension, a result that would also resolve the Kakeya conjecture and Stein’s restriction conjecture. The proof was produced on October 18, 2026, formalized in the Lean proof-checking language the next day, and published on GitHub.

In computer science, OpenAI says its system proved that the Exponential Time Hypothesis (ETH) is false. The ETH holds that certain logic problems — checking whether a set of true-or-false conditions can all be satisfied — cannot be solved much faster than exponential time. Hundreds of papers in graph theory, databases and computational biology build on it. OpenAI says its model found the disproof on October 16, 2026, checked it in Lean the next day, and published both the write-up and the Lean proof on GitHub. It amounts to a subexponential-time algorithm for fixed-width satisfiability problems.

In physics, OpenAI says its models produced a rigorous mathematical existence proof for certain four-dimensional quantum field theories, including quantum chromodynamics (QCD), the theory of the force holding atomic nuclei together. Pure Yang-Mills existence — part of the Clay Mathematics Institute’s Millennium Prize problem on Yang-Mills Existence and Mass Gap — is included as a subproblem. The Yang-Mills proof is dated September 9, 2026, and the QCD proof September 18. Unlike the other two results, the physics materials have not been released: the review group is holding them back and writing the work up itself.

How verification will work

OpenAI is not asking the public to take its word for it. For the computer science claim, Russell Impagliazzo and Ramamohan Paturi of UC San Diego — two of the researchers who originally proposed the ETH — together with Ryan Williams of MIT, will lead an open, independent examination, unpaid and without non-disclosure agreements. They stress their involvement is not certification of the claim. For physics, a group led by Nobel laureate David Gross will shepherd the work, and OpenAI says it is not seeking academic credit for any of the three results.

The details are still contested: the mass gap — the second half of the Millennium Prize problem — is not claimed, and the physics paper itself has not been released for scrutiny. Until the shepherd groups publish verdicts, the honest description of all three results is “claimed, with machine-checked proofs in two cases, awaiting human verification.”

Why it matters for small business

No small business will change a tool or workflow because of this announcement — no new product, price or availability was announced. The story matters because it shows where AI capability is heading: an AI that produces a proof a machine can check is a different kind of coworker from a chatbot that writes fluent text.

The practical lesson applies this week: checkable beats fluent. Before an AI-generated quote, contract summary, line of code or compliance answer reaches a customer, run it through a check you trust — a human review, a second source, a test. If the output cannot be verified, do not ship it.

It also resets timelines: even if all three claims survive review, practical payoffs — faster optimization, new materials, better simulation — are years away, not next quarter’s update. Be skeptical of any vendor selling a “breakthrough-powered” tool on this week’s headlines.

Pricing, availability and what to do next

There is nothing to buy and nothing to sign up for: this was research, not a launch. Watch for the shepherd groups’ verdicts before repeating any claim as fact, and do not reprice, retool or replan on unverified results. If you use optimization or simulation software, ask vendors what actually changed in their product — this month, the answer should be nothing.

Sources

OpenAI announcement, October 26, 2026 (as reported by PANews, Daily AI Brief and AI Weekly); PANews, “OpenAI Announces Three New AI Science Discoveries, Says It Has Disproved the Exponential Time Hypothesis”; Daily AI Brief, “OpenAI Says Its AI Model Refuted the Exponential Time Hypothesis”; AI Weekly, “OpenAI Hands ETH, Kakeya, Yang-Mills Proofs to Reviewers”.