OpenAI Claims Major Math and Physics Proofs; Experts Demand Scrutiny

OpenAI is an artificial intelligence company that on Oct. 26, 2026, announced its internal models produced novel proofs in mathematics, computer science, and physics, including a disproof of the Exponential Time Hypothesis.

OpenAI said on Oct. 26, 2026, that its internal models produced novel proofs in mathematics, computer science, and physics, including a disproof of the Exponential Time Hypothesis, a resolution of the Kakeya conjecture through the local smoothing conjecture, and an existence proof for quantum chromodynamics and pure Yang-Mills theory in four dimensions. The company said it is not seeking academic credit and has formed external expert shepherd groups to verify the results. The announcement drew immediate caution from the original proposers of the Exponential Time Hypothesis, who said their involvement in the review process is not certification of the claim.

The three results span distinct fields. In computational complexity theory, ETH is a conjecture that has been used in hundreds of papers to derive conditional hardness results. In harmonic analysis, a proof of the local smoothing conjecture in every dimension would resolve the Kakeya conjecture and Stein's restriction conjecture. In mathematical physics, an existence proof for quantum chromodynamics and pure Yang-Mills theory in four dimensions would address part of a Clay Mathematics Institute Millennium Prize problem.

OpenAI described the work in a post titled “Working with academic communities to build scientific understanding.” According to the company, its internal models generated the ETH disproof on Oct. 16, 2026, and formalized it in the Lean proof assistant on Oct. 17, 2026. The local smoothing and Kakeya proof was generated on Oct. 18, 2026, and formalized on Oct. 19, 2026. OpenAI said the results will be released under different strategies and reviewed by external expert groups rather than through traditional academic credit.

1. Oct. 16, 2026: An internal OpenAI model generated a proof that would disprove the Exponential Time Hypothesis.
2. Oct. 17, 2026: The ETH disproof was formalized in the Lean proof assistant.
3. Oct. 18, 2026: A proof of the local smoothing conjecture in every dimension was generated.
4. Oct. 19, 2026: The local smoothing and Kakeya proof was formalized in Lean.
5. Oct. 26, 2026: OpenAI publicly announced the three discoveries and the formation of external expert shepherd groups.

Result	Claimed achievement	Generated	Formalized in Lean	Review lead
ETH disproof	Disproof of the Exponential Time Hypothesis	Oct. 16, 2026	Oct. 17, 2026	Russell Impagliazzo, Ramamohan Paturi, Ryan Williams
Local smoothing and Kakeya	Proof of the local smoothing conjecture in every dimension, resolving Kakeya and Stein's restriction conjecture	Oct. 18, 2026	Oct. 19, 2026	External expert shepherd group
QCD and Yang-Mills	Existence proof for QCD and pure Yang-Mills theory in four dimensions, addressing part of a Millennium Prize problem	Not disclosed	Not disclosed	External expert shepherd group

What did OpenAI announce on Oct. 26, 2026?

OpenAI's Oct. 26 announcement describes “three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory,” according to the company's post. The announcement says OpenAI is working with academic communities to build scientific understanding. The company said it is not seeking academic credit and has formed external expert shepherd groups for verification, with different release strategies per result. The announcement did not include full proofs, peer review, or a timeline for completion of the reviews.

The public record of the announcement is consistent with Wikipedia's Exponential time hypothesis article, which states that on Oct. 26, 2026, OpenAI claimed to have disproved ETH. The article also notes that in October 2026, OpenAI and Anthropic both announced disproofs of ETH and that a team of experts announced a collaboration to examine the work. The existence of two claimed disproofs from separate laboratories raises the stakes for independent verification.

Why does the Exponential Time Hypothesis matter?

ETH is a foundational assumption in computational complexity theory. It asserts that 3-SAT, the problem of deciding whether a Boolean formula in conjunctive normal form with three literals per clause is satisfiable, cannot be solved in subexponential time. The conjecture was introduced by Russell Impagliazzo and Ramamohan Paturi and has been widely believed to be true. OpenAI said ETH “has been widely believed to be true, and has been used in hundreds of papers” to derive conditional hardness results.

A disproof of ETH would not merely remove one conjecture from the literature. It would invalidate the conditional conclusions of papers that rely on ETH as an assumption, including work in graph theory, computational biology, databases, and computational geometry. Researchers would need to re-examine which hardness results can be salvaged and which need new assumptions. The review group's statement explicitly calls the result one of significance, which is why the three researchers said careful scrutiny is required.

Who is reviewing the ETH disproof?

OpenAI said the ETH review is led by the original proposers of the conjecture, Russell Impagliazzo and Ramamohan Paturi of UC San Diego, along with Ryan Williams of MIT. The three researchers issued a joint statement as leads of the ETH review group. In that statement, they said, “Our involvement should not be taken as certification of the claim. A result of this significance requires careful, independent scrutiny.” The statement is a direct warning that participation in the review does not imply the proof is valid.

Our involvement should not be taken as certification of the claim. A result of this significance requires careful, independent scrutiny.Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams, Leads of the ETH review group, UC San Diego and MIT

What are the Kakeya and Yang-Mills claims?

OpenAI said its models produced a proof of the local smoothing conjecture in every dimension, which resolves the Kakeya conjecture and Stein's restriction conjecture in all dimensions. The Kakeya conjecture asks how small a set in Euclidean space can be if it contains a line segment in every direction. The local smoothing conjecture is a related harmonic analysis problem about the regularity of solutions to wave equations, and Stein's restriction conjecture concerns the behavior of Fourier transforms on curved surfaces. A proof in every dimension would be a major advance in harmonic analysis.

OpenAI also claimed a rigorous existence proof for quantum chromodynamics, the quantum field theory of the strong nuclear force, and for pure Yang-Mills theory in four dimensions. The Clay Mathematics Institute lists Yang-Mills existence and mass gap as one of its seven Millennium Prize Problems. OpenAI described its result as addressing part of the problem, not as a complete resolution of the full Millennium Prize problem. The claim, if correct, would establish that the theory exists in a mathematically rigorous sense, a necessary step toward the mass gap question.

How will the results be verified?

OpenAI said it is not seeking academic credit and has formed external expert shepherd groups for verification, with different release strategies per result. For the ETH disproof, the review group is led by Impagliazzo, Paturi, and Williams. For the harmonic analysis and quantum field theory results, OpenAI said it has formed external expert shepherd groups, though the announcement did not name all members. The company said the proofs will be shared with these groups before public release.

Part of the verification can be done mechanically. The ETH disproof was formalized in the Lean proof assistant, an interactive theorem prover that checks every step of a formal proof. Formalization is a strong check on logical validity, but it does not by itself confirm that the formal statements match the mathematical conjectures they are meant to prove. Expert reviewers still need to compare the Lean-formalized statements with the standard formulations of ETH, the local smoothing conjecture, and related problems.

No peer-reviewed publication has been announced for any of the three results. The absence of peer review was a central concern in the ETH review group's statement. The group said a result of this significance requires careful, independent scrutiny, and the statement was issued before any certification. The review process is therefore the main check on the claims, rather than traditional journal publication.

What are the market and research implications?

The claims could change the economics of mathematical research. If frontier models can generate proofs of major conjectures, the bottleneck shifts from discovery to verification. That would increase the value of formal proof tools such as Lean and increase demand for mathematicians who can translate informal arguments into formal statements. OpenAI's decision not to seek academic credit also reflects a new pattern in which AI developers release results without traditional authorship.

The broader AI industry is implicated because OpenAI is not the only company making such claims. Wikipedia's Exponential time hypothesis article records that OpenAI and Anthropic both announced disproofs of ETH in October 2026. Two competing disproofs from different laboratories would need to be reconciled, and the outcome could influence how research institutions evaluate AI-generated mathematics. If one or both disproofs fail under scrutiny, the episode would still demonstrate the scale of verification infrastructure needed for AI-era results.

What do the experts say?

The most prominent expert reaction available is the joint statement from the ETH review group. Impagliazzo, Paturi, and Williams said their involvement should not be taken as certification. They described the result as one of significance that requires careful, independent scrutiny. The statement does not confirm or reject the disproof, and it does not describe any technical findings. It is a procedural caution from the researchers best positioned to evaluate the claim.

Wikipedia's article on ETH records that a team of experts announced a collaboration to examine the work, which is consistent with OpenAI's account of external shepherd groups. In the absence of peer review, the mathematical community is likely to demand full proofs, Lean files, and a public explanation of how the formal statements map to the original conjectures. The review process will determine whether the claims become accepted mathematics or remain contested.

What happens next?

OpenAI said it will use different release strategies for each result, and the external expert shepherd groups will decide when and how the proofs are made public. The ETH review group's first task is to verify that the Lean-formalized statement is equivalent to the standard statement of ETH and that the proof is logically sound. For the harmonic analysis and quantum field theory results, the relevant expert groups will need to review the mathematical content and the correspondence between formal and informal statements.

The announcement did not specify a timeline for completion of the reviews. If the claims hold, they would be among the most significant mathematical results produced by AI to date. If they do not, the episode would still be notable for the scale of the verification effort and for the precedent of original conjecturers leading the review of an AI-generated disproof. Either way, the verification phase will be as closely watched as the claims themselves.