By Hamza Faisal

OpenAI says its internal AI models have produced proofs for three problems that have stood unsolved for decades — one each in mathematics, computer science and physics. The company announced the results on October 26, in what it described as a new step from AI systems that answer questions toward systems that make discoveries.

Three results, three fields

In mathematics, OpenAI says its models proved the local smoothing conjecture in all dimensions. The conjecture, proposed in 1991, concerns how waves like light and sound can concentrate at a point without staying concentrated for long. Until now it had been proved only in certain dimensions. If confirmed, the result would also resolve two related problems: the Kakeya conjecture and the Stein restriction conjecture.

In computer science, the company says its system proved that the Exponential Time Hypothesis is false. The hypothesis, proposed in 1999 by Russell Impagliazzo and Ramamohan Paturi, holds that certain problems — most famously Boolean satisfiability, or SAT — cannot be solved much faster than exponential time. Hundreds of papers in graph theory, computational biology, databases and other fields use it to derive limits on how efficiently problems can be solved.

In physics, OpenAI says its models produced a rigorous mathematical existence proof for interacting four-dimensional quantum field theories, including quantum chromodynamics, the theory of quarks and gluons. The work also includes an existence proof for pure Yang-Mills theory, part of one of the seven Millennium Prize problems set by the Clay Mathematics Institute, which carries a $1 million award.

Why the ETH claim matters

The Exponential Time Hypothesis is a load-bearing assumption in theoretical computer science. If it falls, researchers would need to revisit hardness results that treat many problems as effectively unsolvable at scale. MIT professor Ryan Williams, writing on X on October 26, said the claimed proof puts 3SAT on n variables at roughly exp(n/(log n)^{1/2+o(1)}) time — faster than the exponential bound the hypothesis implies.

Notably, the people who proposed the hypothesis will be among those checking whether it is wrong. Williams said he, Impagliazzo and Paturi of the University of California, San Diego, have been asked to organise an open, independent examination of the claim. OpenAI says the mathematics and computer science results have been formally verified using Lean, a tool for checking mathematical proofs, and that the proof materials have been made public.

Verification still pending

The physics result is further from confirmation. Its paper has not yet been released, and the mass-gap part of the Yang-Mills problem is not claimed. A group led by Nobel laureate in physics David Gross will shepherd that work, compiling it before publication. OpenAI says it is not seeking academic credit for the discoveries, and that the external groups will work voluntarily, without non-disclosure agreements, and can publish their own conclusions.

Caution has already followed the announcement. One mathematician noted that researcher Ethan Sussman has been publicly writing up four-dimensional Yang-Mills existence using the same Balaban approach, raising the question of how much is new. And on October 25, OpenAI's own advisory panel on mathematics recommended that AI companies not designate particular people to receive results first, citing unequal access and borrowed reputation.

For now, the three claims sit in the same place most striking AI science claims do: formally checked in part, independently unconfirmed in full. Whether they survive scrutiny will be decided not by the models that produced them, but by the mathematicians and physicists now poring over the proofs.