OpenAI puts three AI-generated science claims before researchers

A conjecture about waves. An assumption that shapes computer science. A mathematical foundation for particle theories. OpenAI’s October 26 announcement puts AI-generated arguments on all three before researchers, but the available evidence differs sharply from one claim to the next. The mathematics and computing materials are public. The physics artifacts are not. In the sources reviewed for this article, outside specialists have not yet delivered a completed, independent assessment of the new claims.

What the mathematics claim says

The first result concerns local smoothing: broadly, whether a wave’s sharpest focusing becomes less extreme when measured across time rather than at a single instant. OpenAI’s public description of its Lean formalization specifies a wave estimate for every spatial dimension of at least two. It also lists consequences involving Stein’s restriction problem and Kakeya sets, objects that contain a line segment pointing in every direction.

The Kakeya claim is precise. The formalization’s scope document says these sets have full Hausdorff dimension in every dimension. That measures their fractal size; it does not say they must occupy positive volume. The same document records several related estimates, so reducing the work to a single geometric slogan would miss much of what is being claimed.

A natural-language manuscript and formal proof materials are posted for examination. OpenAI says the AI output arrived October 18 and was formalized in Lean the following day. A Lean check can establish that a formal proof follows from its stated definitions and assumptions. Mathematicians still need to assess how faithfully that formal statement captures the intended theorem, whether the reasoning and dependencies withstand scrutiny, and which ideas are new.

Why a computing conjecture matters

The Exponential Time Hypothesis, or ETH, proposes a strong lower bound on the time needed to solve certain Boolean satisfiability problems as their number of variables grows. Many theoretical results use it as a conditional starting point. If the relevant formulas can instead be handled in subexponential time, those conditional conclusions need to be reconsidered.

OpenAI’s Lean scope document describes a deterministic machine that exactly counts satisfying assignments for a specified encoding of fixed-width Boolean formulas in subexponential time. It distinguishes this formalized bound from a sharper bound in the October 25 manuscript. The distinction matters: a headline about ETH has to be traced back to the exact algorithm, encoding and time bound actually established.

Even if the argument holds, it is not a P-versus-NP solution. Subexponential time can still become prohibitively large, and a theoretical bound is not a demonstrated improvement in today’s software. OpenAI dates the AI-generated argument to October 16 and its Lean formalization to October 17. An outside group is organizing its examination, but its participation is not a completed endorsement.

What the physics claim does—and does not—cover

OpenAI claims existence proofs for certain four-dimensional quantum field theories, naming quantum chromodynamics, a proposed completion of the Standard Model and pure Yang–Mills theory. These are mathematical claims, not reports of newly observed particles or experimental confirmation.

There is a crucial boundary. The Clay Mathematics Institute’s Yang–Mills problem demands existence and a mass gap—an additional property of the theory’s energy spectrum. OpenAI claims existence, not the mass gap or the full Millennium Prize solution. It says a Lean formalization of Yang–Mills existence was completed October 16, but the AI-produced physics materials are being withheld while outside researchers prepare a write-up. Readers cannot yet inspect the detailed physics argument.

The test ahead

This is a distinct October 26 disclosure, not simply the October 6 collection under a new headline. Yet the public mathematics and computing artifacts appear in OpenAI’s existing math repository; the sources reviewed here do not establish that the work is wholly independent of that earlier corpus.

The useful next updates will be concrete: independent checks of the public formalizations, expert comparisons between those statements and the advertised claims, explanations of what is genuinely new, and release of enough physics detail for wider examination. Those steps may reveal major advances, narrower results or needed corrections. For now, two arguments are open to examination; the third awaits a public account.

Bright has not reproduced the Lean checks or audited the underlying proofs; the physics proof artifacts are not public. This source-based report is not a correctness or originality verdict.