OpenAI says one of its internal models has disproved the Exponential Time Hypothesis. The Next Web reported the claim on October 26, 2026. It is a useful test case for AI proof verification, because the claim is big and the checking has only just begun.
What OpenAI announced
According to The Next Web, OpenAI announced three AI-produced results on Monday, October 26. The first is the claimed refutation of the Exponential Time Hypothesis, a widely believed idea about how hard Boolean satisfiability is to solve. The second is a proof of the local smoothing conjecture in every dimension. The third is a proof that pure Yang-Mills theory exists, a result tied to a Clay Millennium Prize problem.
The report says the model found the ETH disproof on October 16. OpenAI then checked it in Lean, a language that verifies proofs by machine, on October 17. OpenAI published the write-up and the Lean proof on GitHub, the report says.
Who is checking it
Three computer scientists will lead an open, independent check: Russell Impagliazzo, Ramamohan Paturi and Ryan Williams. In their open letter they wrote: "Our involvement should not be taken as certification of the claim." Williams said the group signed no non-disclosure agreements and will not be paid.
A separate physics group led by Nobel laureate David Gross has not released the AI-produced material. The report says that group will write up the work itself. AI Weekly says OpenAI acknowledges that all three claims face months of independent peer review.
What it means for people who check data
A machine-checked proof is strong evidence that the steps hold. The Next Web does not say whether the formal statement matches the original conjecture. That question is part of the review still under way. The same gap exists in business work. A passing check only covers what it was built to test.
Three habits carry over to data verification:
- Treat a vendor announcement as a claim until someone independent confirms it.
- Ask what the automated check actually covers.
- Wait for raw data and open review before acting on a result.
For routine reports, our guide to data validation explains how format checks differ from truth checks. The trust index shows how we weigh evidence that has not yet been reviewed.
The claims remain OpenAI's. Nothing here has been confirmed by outside reviewers yet.