OpenAI Claims AI Proofs Upend ETH, Kakeya, and QCD Existence
By @sharedot · · 8 pages
- AI Frontier
- Openai
- Mathematics
- Complexity Theory
- Quantum Field Theory
OpenAI claims AI models produced unverified proofs disproving ETH, settling Kakeya in all dimensions, and proving QCD exists.
Three claimed breakthroughs in one drop
On October 26, 2026, OpenAI announced that its internal models produced three research results spanning mathematics, computer science, and physics: a proof of Sogge's local smoothing conjecture in every spatial dimension of at least two, a disproof of the Exponential Time Hypothesis, and a mathematically rigorous existence proof for interacting four-dimensional quantum field theories including QCD.
The company frames the AI-generated manuscripts as 'raw research data' that must still be converted into scientific understanding, and it is testing a new release model to get there. All three results remain unconfirmed by the scientific community.
Why the ETH reversal is the upset
The computer science paper carries the boldest claim: the Exponential Time Hypothesis, widely believed true since Impagliazzo and Paturi proposed it in 1999, is false.
OpenAI says its system produced a deterministic algorithm that exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space — the 3-SAT case refuting ETH.
As Firstpost notes, ETH underpins hundreds of conditional hardness results across graph theory, computational biology, databases, and geometry, so a confirmed refutation would force researchers to reconsider results that rely on it.
The evidence: Lean proofs on GitHub
According to unite.ai, the ETH disproof was produced in natural language on October 16, 2026 and formalized in Lean on October 17, while the local smoothing output was obtained October 18 and formalized the following day.
The AI write-up and Lean proof are public in OpenAI's GitHub repository, and at the verification group's request all raw data was released for independent checking.
Lean verification adds weight, but AI Insider points out the limits: a proof assistant confirms the formal argument, not that the formal statement captures the intended problem — that match is what experts must check.
Shepherds, not peer review
OpenAI recruited external experts as 'shepherds' who chose three different release strategies: the math output went out directly, a computer science group led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams published alongside the raw output, and a physics group led by Nobel laureate David Gross with Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov deferred releasing materials until it writes up the result itself.
Collaborations run under four principles — transparency, no academic credit for OpenAI, recognition for explainers, and independence — with participation voluntary and unpaid, unite.ai reports.
The physics claim: QCD made rigorous
The hardest result to assess is a claimed existence proof for interacting quantum field theories with gauge fields and fermions in four dimensions, including quantum chromodynamics and a unified completion of the Standard Model, built on Balaban's constructive approach.
According to GIGAZINE, the models produced a natural-language proof of pure Yang–Mills existence on September 9, 2026 and QCD on September 18, with Lean formalization of Yang–Mills existence completed October 16.
AI Insider cautions an existence proof alone addresses only part of the Clay Institute's Yang–Mills Millennium Prize problem, which also requires a mass gap.
Stakes and what comes next
OpenAI admits the approach has tradeoffs — chosen shepherds create information asymmetries — and its own AGMAI math advisory panel recommended against giving designated people privileged early access, favoring scholarly repositories, Lean proofs, and funded explanation work, per The Verge.
The computer science group published an open letter asking how credit should recognize verification work and how to support students as methods shift.
OpenAI has committed to funding a conference on the ETH implications and continues its $5 million Simons Foundation contribution to help researchers understand AI-generated results.
Sources
- theverge.com › OpenAI announces three big science discoveries and a new way to work with academics
- unite.ai › OpenAI Says Its Models Produced Three Proofs in Math, CS, and Physics
- firstpost.com › OpenAI claims three new AI breakthroughs in math, physics and computer science
- theaiinsider.tech › OpenAI Opens Three AI Research Claims to Academic Examination
- gigazine.net › OpenAI has announced three new findings in mathematics, computer science, and physics