On October 26, 2026, OpenAI announced three new scientific results produced by its internal artificial intelligence models. The results concern mathematics, computer science, and physics, and together with them the company presented a new way of working with academic communities. In this approach, outside experts are brought in as shepherds — specialists who follow the results through the review process and evaluate them. OpenAI itself states openly that these results have not yet undergone independent peer review. This article therefore presents the scientific results not as confirmed facts, but as claims by OpenAI. The company also said it is not seeking academic credit — recognition of scientific authorship. This position is one of the most important aspects of the announcement, because it leaves the task of evaluating the results entirely to the academic community. All three results were announced on a single day, as one package, but each will follow a separate review path. The mathematics result has been made open for researchers to evaluate, while the computer science and physics results are awaiting independent review. The announcement itself is thus a single event, but its scientific fate splits into three different processes. OpenAI's statement was published on the company's official blog, and the story was also covered by The Verge. Both sources confirm the same core facts: the announcement date, the three fields, the shepherd model, and the fact that independent review has not yet been completed.
Mathematics: The Local Smoothing Conjecture
According to OpenAI's mathematics claim, one of its models has proved the local smoothing conjecture in all dimensions. The company asserts that this proof also resolves the Kakeya conjecture and the Stein restriction conjecture. These three problems are described as connected, and the local smoothing result is claimed to provide a solution to the other two as well. The proof was prepared on October 18, 2026, and the next day, October 19, it was formalized in Lean. Formalization in Lean means the proof has been put into a machine-checkable form, a step singled out in the announcement. The mathematics result differs from the other two in this respect: it has been made open for researchers to evaluate. That is, the mathematics community can already study, check, and assess the proof. This openness is a different path from the shepherd model repeated in the rest of the announcement, and it means the evaluation process for mathematics has already begun. At the same time, being openly available still does not mean the result has passed independent peer review. So far, the story is only about OpenAI's claim and its formalization in Lean. The final scientific judgment will come from the mathematics community's review. If the review confirms the claim, three connected problems will have been solved at once; if it does not, the claim will count as rejected. In either case, what is clear at this stage is one thing: a proof is claimed to exist, it has been formalized in Lean, and it is open for evaluation.
Computer Science: The Exponential Time Hypothesis
For computer science, OpenAI claims that the Exponential Time Hypothesis, known as ETH, is false. The company asserts that its system has proved ETH false, and that this result was subsequently checked in Lean. ETH is described as one of the central assumptions of theoretical computer science, which is why the claim of its refutation has attracted so much attention. Unlike the mathematics result, however, the computer science claim is still awaiting independent review. The independent review will be carried out by a group led by Russell Impagliazzo and Ramamohan Paturi of the University of California, San Diego, and Ryan Williams of the Massachusetts Institute of Technology. This group will study and evaluate the claim independently of OpenAI. The fact that the group's members are named in the announcement shows the transparency of the review process: it is known in advance who will do the checking. Until the review is complete, the ETH claim also does not count as a confirmed scientific result. OpenAI itself is not seeking academic credit for this result either — that is, it is not demanding recognition of authorship of the claim. This approach appears aimed at avoiding pressure on the review group's work. What happens next in computer science now depends precisely on that group's conclusion. If the group confirms the claim, the falsity of ETH will become a result accepted by the scientific community. If the group finds flaws in the claim, the result will be revised or rejected. At this stage, the definite facts for the reader are these: the claim has been announced, it has been checked in Lean, the independent review group has been named, and its conclusion is awaited.
Physics: Interacting Four-Dimensional Quantum Field Theories
For physics, OpenAI claims that a rigorous proof has been given for the existence of interacting four-dimensional quantum field theories. Such theories include quantum chromodynamics. The company asserts that one of its models has produced a rigorous proof of the existence of these theories. Within this claim there is another important part: the problem of the existence of pure Yang-Mills theory is part of one of the Clay Millennium Prize problems. In other words, the physics claim is not only a single result, but also touches a part of the long-standing list of prize problems. The physics result is also awaiting independent review, and in this process a group led by Nobel laureate David Gross is taking part as shepherd. The Gross group will follow the physics claim and evaluate it. This choice is singled out in the announcement, because the shepherd's scientific standing determines trust in the review process. Unlike the mathematics result, the announcement does not say that the physics proof has been made open for researchers to evaluate; instead, it stresses that independent review is awaited. The final judgment on physics therefore depends on the work of the Gross group. So far, the definite facts are: the claim has been announced, it concerns the existence of theories including quantum chromodynamics, the pure Yang-Mills part is a portion of a Clay Millennium Prize problem, and the review is being conducted by a group led by David Gross. Beyond these facts, there is as yet no confirmed scientific conclusion.
The Shepherd Model: A New Way of Working With Academic Communities
Besides the three results, another important part of the announcement is the new way of working with academic communities. OpenAI presented a model of bringing in outside experts as shepherds. In this model, outside experts follow the results through the review process, evaluate them, and deliver their conclusions. The Impagliazzo, Paturi, and Williams group for computer science and the Gross group for physics perform exactly this role. For mathematics, the process is proceeding differently: the result is directly open for researchers to evaluate. The shepherd model and the open-evaluation path complement each other: one implies review through named expert groups, the other through the wider community. The fact that OpenAI is not seeking academic credit is also part of this model. The company does not demand recognition as the author of the claims, leaving the right to evaluate and conclude to the academic community. This approach amounts to testing a new practice on the question of authorship of scientific results obtained with artificial intelligence. The model itself is also still at the trial stage: its effectiveness will be judged after the review groups' conclusions appear. If the groups confirm the claims, the shepherd model will be seen as a working example of cooperation between artificial intelligence and academia. If the claims are rejected, the model itself will also face critical assessment. In either case, the current announcement signals the launch of the model — not its final outcome, but its beginning.
What Has Not Yet Been Confirmed
To read this announcement correctly, it is important to separate what is confirmed from what is claimed. Confirmed facts: OpenAI announced three results on October 26, 2026; the mathematics proof was prepared on October 18 and formalized in Lean on October 19; independent review groups for computer science and physics have been named; the company is not seeking academic credit. Claimed but not yet confirmed: that the local smoothing conjecture has been proved in all dimensions and resolves the Kakeya and Stein restriction conjectures, that the Exponential Time Hypothesis is false, and that there is a rigorous proof of the existence of interacting four-dimensional quantum field theories, including quantum chromodynamics. This distinction defines the essence of the announcement and must be preserved in any retelling. The following statement expresses OpenAI's own position:
The results have not yet undergone independent peer review: OpenAI is not seeking academic credit, the mathematics result is open for researchers to evaluate, and the computer science and physics claims are awaiting review by independent expert groups. — OpenAI statement, via The Verge and the company's official blog
The next stage of this announcement now depends not on OpenAI, but on the reviewers. An assessment for mathematics is expected from the broad researcher community, for computer science from the Impagliazzo, Paturi, and Williams group, and for physics from the group led by David Gross. Until their conclusions appear, all three results remain claims. That is why, in reading this news, the most important word is "claim": OpenAI is claiming, and the scientific community is checking. When the review conclusions appear, the story will move to a new stage, and it will then become clear whether the claims become confirmed results or are rejected. When the review conclusions are published, each claim will be assessed separately, because the fates of the three results will be decided independently of one another.
