A three-front scientific claim
OpenAI has announced three claimed discoveries spanning harmonic analysis, computational complexity and quantum field theory, presenting them not as finished textbook results but as AI-produced research artifacts that now need expert interpretation . The working headline matches the substance of the announcement: OpenAI says its models have generated major proofs in mathematics, computing and quantum physics, including results tied to the local smoothing conjecture, the Exponential Time Hypothesis and the existence of quantum chromodynamics .
The scale of the claim is striking because the three areas normally belong to different expert communities. Harmonic analysis studies waves, oscillation and concentration; complexity theory studies what computers can or cannot do efficiently; quantum field theory gives the mathematical language behind particle physics . OpenAI says it asked external specialists to serve as “shepherds” for the results, with different release strategies depending on the field: immediate public release for the mathematics output, immediate public release plus a structured review group for the computer science output, and deferred release for the physics materials while an expert group prepares its own account .
That framing matters. The company is not saying that journal refereeing is complete, nor that the mathematical and physical consequences are already settled by consensus. OpenAI’s own announcement describes the manuscripts as raw research data that must be converted into scientific understanding by academic communities . Al Jazeera reported the same cautious posture: outside scientists are being asked to verify, examine and explain the work, and at least one prominent complexity theorist said his group is still determining whether the ETH proof holds up .
The mathematics result: local smoothing, Kakeya and restriction
The mathematics claim concerns the local smoothing conjecture, proposed by Christopher Sogge in 1991, which addresses how wave-like phenomena concentrate over time . In plain terms, the conjecture says that the most extreme focusing of waves should be short-lived, and that averaging over time produces smoother behavior than the worst instant might suggest . OpenAI says its model produced a proof in every spatial dimension, and that this result also resolves the Kakeya conjecture and Stein’s restriction conjecture in all dimensions .
The announcement places the claim in a precise historical sequence. OpenAI notes that Larry Guth, Hong Wang and Ruixiang Zhang proved the local smoothing conjecture in two spatial dimensions in 2019, while Wang and Zahl proved the three-dimensional Kakeya conjecture in 2025 . Interesting Engineering similarly described the new claim as a proposed leap from important partial results to all dimensions, adding that OpenAI obtained the new output on October 18, 2026, and formalized it in Lean the next day .
The release strategy for this result is the most open of the three. OpenAI says it has made both the AI-generated natural-language write-up and a Lean proof available in its public GitHub repository . Lean is important because it can check formal proof steps mechanically, but it is not the same thing as the whole mathematical community understanding why a proof works, how it relates to earlier methods, or whether every informal consequence has been stated in the strongest legitimate form. Quartz summarized the distinction well: the local smoothing output was formalized, posted, and then left for researchers to examine, while the broader interpretation still depends on human experts .
Reaction has already begun. Alex Iosevich, a University of Rochester mathematician who works in harmonic analysis, told Al Jazeera that OpenAI appeared to have “cracked” the conjecture and called the development historic . His reaction captures both the excitement and the disruption: if the proof survives scrutiny, it would not merely add one theorem to a list but reshape a cluster of problems that has guided parts of analysis for decades.
The computing result: a claimed disproof of ETH
The second claim may be even more destabilizing for theoretical computer science. OpenAI says its system proved that the Exponential Time Hypothesis is false, and that the result was subsequently verified in Lean . ETH is a widely used conjecture about the hardness of Boolean satisfiability problems, closely connected to the study of P versus NP and to the lower-bound assumptions used across algorithmic research .
According to OpenAI, ETH has been used in hundreds of papers to derive conditional hardness results in graph theory, computational biology, databases, computational geometry and other fields . If the refutation is accepted, many of those papers would not simply become useless; rather, their conditional structure would have to be reinterpreted because the assumption on which they rely would be false. The Economic Times reported OpenAI’s position that the claimed finding “reopens questions” about how much faster certain tasks could become, while also noting that practical improvements would require significant further research .
OpenAI says the natural-language disproof was found on October 16, 2026, and formalized in Lean on October 17, 2026 . The company has named Russell Impagliazzo and Ramamohan Paturi of UC San Diego, along with Ryan Williams of MIT, as leaders of the group that will help verify and shepherd the computer science result . That roster is notable because Impagliazzo and Paturi were among the original proposers of ETH, while Williams has worked extensively on related algorithmic and lower-bound questions .
Williams told Al Jazeera that a confirmed refutation would be a major development because a great deal of theoretical computer science has been built on ETH being true since the late 1990s . He also emphasized that his group is still verifying and examining the proof . This is the crucial distinction for readers outside the field: a Lean-checked formalization is powerful evidence about a formal statement, but the community still has to inspect the definitions, the scope of the formal theorem, the translation from informal claim to machine-checked object, and the consequences for existing literature.
OpenAI says it has agreed to the group’s request to make the raw data public and to fund a conference on the implications of the work . In practice, that means the next phase is likely to look less like a single announcement and more like a distributed audit: seminars, blog posts, formal proof inspections, independent reproductions and possibly competing explanations of what the new algorithmic method actually does.
The physics result: QCD, Yang–Mills and a slower release
The third claim moves into quantum field theory, where OpenAI says its models produced a rigorous existence proof for certain four-dimensional interacting quantum field theories with gauge fields and fermions, including quantum chromodynamics, or QCD . QCD is the theory of the strong interaction between quarks and gluons, and OpenAI says the work also includes a proof of pure Yang–Mills theory, which is part of a Clay Mathematics Institute Millennium Prize problem .
This result is being handled more cautiously than the other two. OpenAI says David Gross will lead a group including Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov to shepherd the work . The company also says the group decided to hold off on releasing the AI-produced physics artifacts and instead prepare its own account first . Quartz reported the same release posture, noting that the physics group is delaying publication of the underlying materials while it writes up the work .
The timeline is unusual. OpenAI says its models produced a natural-language proof for Yang–Mills existence on September 9, 2026, and a proof for QCD on September 18, 2026; the company says it finished formalization of Yang–Mills existence on October 16, 2026 . The QCD claim is therefore not yet available to the public in the same way as the mathematics and ETH results. The result may be momentous, but at the current stage the broader community is being asked to wait for the expert group’s explanation before assessing the details.
That caution is justified by the stakes. A rigorous four-dimensional construction related to Yang–Mills and QCD would sit at the boundary between mathematics and theoretical physics, where physical intuition, renormalization methods and formal existence questions have long been difficult to align. OpenAI says its work builds on Balaban’s approach in constructive quantum field theory . If the proof survives review, it could change how physicists and mathematicians discuss the foundations of quantum field theory; if it does not, it may still become an important case study in how AI-generated arguments should be audited.
Why verification is now the story
The central news is not simply that OpenAI made three claims. It is that the company has exposed a new model of scientific publication in which AI systems generate possible breakthroughs, formal methods check selected statements, and human communities must decide how to turn artifacts into shared knowledge . OpenAI says the collaborations will follow principles of transparency, no academic credit for the company, recognition for the experts who explain the work, and independence for participants .
The current state is therefore mixed: the mathematics and computing materials are public and formalized; the physics materials are not yet public in full; and outside validation is still under way . That does not make the announcement insignificant. It makes it a live scientific event rather than a completed chapter.
If even one of the three claims withstands scrutiny, it would be a major result. If all three do, the announcement would mark a turning point in the relationship between AI systems and frontier science. For now, the right verb is still “claims,” not “settles.” But the claims are specific, recent, partially formalized and already in the hands of experts who can test them.