The Pulse
OpenAI Announces Three AI-Generated Science Results
OpenAI announced results in harmonic analysis, computational complexity and quantum field theory on October 26, 2026. The company released two results for examination and says outside researchers will help review them, while a physics resul
AI.info Team · Written with AI · fact-checked ·
OpenAI releases three results under three different rules
On October 26, OpenAI announced three results produced by its internal models: a proof concerning the Kakeya conjecture, a claimed refutation of the Exponential Time Hypothesis, and a result on the mathematical existence of quantum field theories including quantum chromodynamics. The company is not presenting them as finished academic papers. It describes the manuscripts as “a form of raw research data” and says it is testing different ways to bring researchers in to examine and explain the work.
“We do not yet know which of these three approaches will be most useful, and we welcome feedback on what works and what we can improve,” OpenAI writes. The release assigns each result a different path: the mathematics output is public, computer scientists will shepherd the complexity result while its materials are public, and the physics group will write up its result before OpenAI releases the model’s artifacts.
OpenAI says the outside collaborations are voluntary and unpaid, and participants will not sign nondisclosure agreements. The company says it is not seeking academic credit for discoveries made by its models, while the groups should receive recognition for explanations and materials they produce. That division matters: OpenAI is making the claims, while the researchers it names are being asked to assess and contextualize them.
The mathematics claim reaches beyond Kakeya
The mathematics result concerns local smoothing, a problem in harmonic analysis about how strongly waves can focus over time. OpenAI says its model produced a proof of the local smoothing conjecture in every dimension, extending earlier results in two and three dimensions. The company says the proof also resolves the Kakeya and Stein restriction conjectures in all dimensions.
OpenAI says it obtained the model output on October 18 and formalized it in Lean the following day. It has published both a natural-language write-up and a Lean proof in its public repository. A Lean formalization checks that a proof follows from its stated formal assumptions; it does not, by itself, establish that the formal statement captures the full mathematical claim or that the argument has been independently vetted.
Computer scientists will examine the ETH refutation
The second result challenges the Exponential Time Hypothesis, a widely used conjecture about the difficulty of solving Boolean satisfiability problems. OpenAI says the conjecture has underpinned conditional hardness results across fields including graph theory, computational biology, databases and computational geometry. If the claim holds, researchers would need to revisit some of those assumptions and consider what faster algorithms may be possible.
OpenAI says its system produced a natural-language disproof on October 16 and that the work was formalized in Lean on October 17. It has published the AI-generated write-up and proof, and says it will make the raw data public. A group led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego and Ryan Williams of MIT will examine and explain the result; OpenAI also agreed to fund a conference on its implications.
The company’s own repository describes the Lean formalization in terms of exact counting for canonically encoded k-CNF formulas under a particular subexponential-time bound. That technical statement is what the formal proof encodes; the outside group’s work will help establish how the result should be interpreted in relation to the broader ETH claim.
The quantum-field-theory result remains under wraps
The physics claim is the most consequential and the least available for public inspection so far. OpenAI says the work gives a rigorous existence proof for certain interacting four-dimensional quantum field theories containing gauge fields and fermions, including quantum chromodynamics and a unified completion of the Standard Model. It also says the work includes a proof of pure Yang–Mills theory’s existence, a central part of a Clay Mathematics Institute Millennium Prize problem.
OpenAI says its models produced a natural-language proof for Yang–Mills existence on September 9 and a proof for quantum chromodynamics on September 18. The company finished formalizing the Yang–Mills result in Lean on October 16, but it has not released those AI-produced materials: the physics group, led by David Gross and including Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov, chose to prepare its own write-up first. OpenAI’s announcement is a claim about a potential proof, not confirmation that the Millennium Prize problem has been accepted as solved.
Verification now depends on what others can inspect
The release makes independent scrutiny central to the story, but the three projects begin from different positions. The math result has public artifacts for researchers to assess; the ETH work has both public artifacts and a named external group; the physics claim awaits a write-up from its collaborators. OpenAI says it will fund researchers to examine, explain and build on the work, alongside a previously announced $5 million contribution to the Simons Foundation for workshops and related programs.
OpenAI’s announcement does not settle whether the proofs withstand expert examination. It sets out what the company says its models found, what material it has made available, and who will take part in the next stage of review. For now, the clearest dividing line is between the two sets of public proofs and the quantum-field-theory artifacts that remain withheld.