Skip to content

AI

OpenAI says its models refuted the Exponential Time Hypothesis alongside two more results

OpenAI also said its models proved the local smoothing conjecture in every dimension and completed a unified version of the Standard Model.

https://breach-signal.com/artificial-intelligence/openai-ai-models-refute-exponential-time-hypothesis/

Share

Key facts

  • OpenAI claimed its internal model refuted the Exponential Time Hypothesis.
  • OpenAI published the ETH write-up and Lean proof on GitHub.
  • Russell Impagliazzo and Ramamohan Paturi of UC San Diego and Ryan Williams of MIT will examine and explain the ETH claim.
  • The three researchers said their involvement is not an endorsement and the result needs independent scrutiny.

OpenAI said on Oct. 26, 2026, that its internal AI model refuted the Exponential Time Hypothesis, which the company described as widely believed to be true, and it published the write-up and Lean proof on GitHub. The company said the model found the disproof on Oct. 16 and checked it in Lean, software that checks mathematical proofs, the following day. The announcement bundled a mathematics result and a physics result alongside the Exponential Time Hypothesis claim.

The hypothesis deals with Boolean satisfiability and connects to P versus NP, and OpenAI said it underlies hundreds of computer science papers.

OpenAI said its models had established the local smoothing conjecture across all dimensions. Christopher Sogge proposed the conjecture in 1991, mathematicians proved the two-dimensional case in 2019, and a 2025 proof settled the three-dimensional Kakeya conjecture. OpenAI said settling the Kakeya conjecture and Stein’s restriction conjecture in every dimension would follow from a proof covering all dimensions.

On Oct. 26, OpenAI said its models had produced rigorous proofs showing that multiple four-dimensional quantum field theories exist. On Oct. 27, the company said its models had completed a unified version of the Standard Model.

Quantum chromodynamics describes the strong force binding quarks and gluons, and Pure Yang-Mills theory is included in a Clay Mathematics Institute Millennium Prize problem. In September 2026, OpenAI said it would refuse the Millennium Prize connected with its Navier-Stokes proof. On Oct. 27, reviewers were concerned that Lean had confirmed a weaker claim than the one OpenAI originally wrote in its Navier-Stokes solution.

On Oct. 26, OpenAI said Nobel laureate David Gross would lead the physics review group, with Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov. Russell Impagliazzo and Ramamohan Paturi of UC San Diego, along with Ryan Williams of MIT, will examine and explain the Exponential Time Hypothesis claim. The physics group said it would not release the AI-produced material immediately and would write up the research itself.

Impagliazzo, Paturi and Williams said their participation does not certify the claim. On Oct. 26, the three researchers said their involvement should not be seen as an endorsement and that the result requires independent scrutiny. Ryan Williams said the group signed no non-disclosure agreements and would receive no payment. The group asked OpenAI to release all raw Exponential Time Hypothesis data and to fund a community conference, and the company agreed.

On Oct. 26, OpenAI said the researchers in each group ought to receive recognition for writing the explanations, and the company said it is not seeking academic credit for discoveries its internal models make. The three computer scientists questioned how credit should be assigned when artificial intelligence produces results first, and they warned that publishers and people with powerful tools might receive credit while verification and explanation get less recognition. Mathematicians had raised comparable concerns in the Leiden declaration of June 2026.

On Oct. 27, OpenAI characterized model-generated texts as raw research data that communities could turn into scientific understanding, and it said it had contacted outside experts to serve as shepherds translating AI outputs into human knowledge. The company acknowledged that picking the first shepherds determines who joins early and produces an information imbalance. Gizmodo reported that an advisory group with past ties to OpenAI criticized its reliance on unreleased internal AI models for breakthrough results.

TNW said OpenAI had released 722 math papers on Oct. 6, a batch the company’s Exponential Time Hypothesis release mirrored, while Gizmodo reported that OpenAI put 377 math results onto GitHub early in October. On Oct. 26, the company contributed $5 million to the Simons Foundation to fund workshops and conferences on AI-produced results.

Was this helpful?

Join the discussion

Please read our Comment Policy before commenting. First-time comments are held for moderation. By posting a comment you agree to our Terms of Use and Privacy Policy, which explain how your name, email address and comment are used.

Your email address will not be published. Required fields are marked *