OpenAI announced on October 26, 2026 that its internal models produced three major research results — a proof of the local smoothing conjecture, a disproof of the Exponential Time Hypothesis, and an existence proof covering quantum chromodynamics — and handed all three to outside academics to verify. Why it matters: these are claims, not settled discoveries, and how the verification goes will say a lot about whether AI-generated science can be trusted.
October 27, 2026
What OpenAI is claiming
The first claim is in harmonic analysis: a proof of Sogge’s local smoothing conjecture, proposed in 1991, in every dimension. OpenAI says the result also implies versions of the Kakeya and Stein restriction conjectures. The company says the output was obtained on October 18 and formalized in Lean, a machine-checkable proof language, the following day.
The second claim targets computer science. The Exponential Time Hypothesis (ETH) is a widely used assumption that certain satisfiability problems cannot be solved in subexponential time; hundreds of papers build conditional hardness results on it. OpenAI says its model found a disproof on October 16, formalized in Lean on October 17, describing a deterministic method that counts satisfying assignments for fixed-width formulas in subexponential time.
The third claim is in physics: existence proofs for four-dimensional interacting quantum field theories, including QCD and pure Yang–Mills — the latter being one component of the Clay Mathematics Institute’s Millennium Prize problem. OpenAI is clear, and independent trackers stress, that an existence proof alone does not settle the full Millennium problem, which also requires a mass gap.
How the checking will work
OpenAI describes the manuscripts as raw research data rather than finished science, and says it does not seek academic credit. The math and computer-science outputs are public, including Lean materials. The physics artifacts are being withheld for now at the request of the expert group reviewing them, so that group can write up the work itself — a delay the Institute for Advanced Study’s advisory group on mathematics and AI has publicly questioned.
The ETH review will be led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego — two of the hypothesis’s original formulators — with Ryan Williams of MIT. The physics work will be shepherded by a group led by Nobel laureate David Gross, with Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov. OpenAI also says it is contributing $5 million to a Simons Foundation initiative supporting workshops and programs on AI-generated mathematics.
Why it matters
Treat the headlines carefully. A Lean check confirms a formal derivation follows its steps; it does not by itself prove the formal statement matches the headline claim, and independent trackers list the ETH result as an unverified claim. Even OpenAI says turning a disproved ETH into faster practical algorithms would take significant further research. The real story this week is the process: an AI lab publishing raw, machine-generated proofs and paying for the human capacity to check them. If the shepherds confirm even one of the three, it is a genuine milestone. If they do not, it is a useful lesson in how to read the next announcement.
Frequently asked questions
Did an AI really disprove the Exponential Time Hypothesis?
OpenAI says so, and says the result was formalized in Lean on October 17. A working group led by two of the hypothesis’s originators is now reviewing it. Until they report, it is a claim under review.
Does the physics result solve the Yang–Mills Millennium Prize problem?
No. OpenAI claims an existence proof for pure Yang–Mills theory. The Millennium problem also requires proving a mass gap, which is a separate element.
Can I read the proofs?
The mathematics and computer-science materials are public. The physics outputs have not been released yet; the reviewing group asked to write up that work first.
Sources: OpenAI announcement (via Pivot News and MLQ News), TechCrunch/Superhuman reporting