All stories
OpenAI October science research review Debates Developing

OpenAI claims an ETH refutation and opens proofs to outside scrutiny

A proposed refutation of the Exponential Time Hypothesis leads a three-result research announcement, with public proof artifacts and outside researchers examining the claims.

Sources (2)

Why now OpenAI published the announcement on October 26 and described public artifacts and external review arrangements; the claims are still being presented for scientific scrutiny.

OpenAI says it has produced a refutation of the Exponential Time Hypothesis, alongside proposed results in harmonic analysis and quantum field theory. Its October 26 announcement places the claims inside a new process for academic scrutiny, describing the manuscripts as raw research data rather than presenting their publication as the end of scientific review.

The computational-complexity claim is especially consequential: ETH concerns the time required to solve satisfiability problems. OpenAI’s public Lean scope document describes a subexponential-time result for exactly counting satisfying assignments of fixed-width Boolean formulas. The company says a natural-language proof was produced on October 16 and formalized in Lean on October 17. These are the company’s account and public proof artifacts, not an independent confirmation by AITrending that the claimed refutation is correct.

OpenAI says an external group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams will help verify and explain the computer-science result, while inviting community examination of the public material. The announcement outlines different review arrangements for the three research areas: public mathematical proofs, open computer-science collaboration, and a physics group examining artifacts initially held back from public release.

The immediate news is therefore both the scale of the claims and how they will be tested outside the company. The announcement does not establish that P equals NP, nor does it promise practical fast SAT solvers today. OpenAI itself says practical algorithms require further research. Acceptance will depend on researchers checking the proofs, their assumptions and the interpretation of the formal statements.