OpenAI says it has made substantial progress on the Hodge conjecture—but has shown no proof

OpenAI says it has made “substantial progress” on another Millennium Prize problem. The Information identifies it as the Hodge conjecture, a central question in algebraic geometry about whether certain topological features of a smooth complex projective variety arise from algebraic subvarieties.

The claim is deliberately short of “solved.” No proof has been released, no independent mathematicians have verified it and the company says it is working out how to share the result. Those qualifications are decisive: advanced mathematics is not validated by a model’s confidence or an internal benchmark but by a public argument that specialists can inspect line by line.

The disclosure follows OpenAI’s claimed Navier–Stokes breakthrough using thousands of agents. That episode became contentious after mathematicians working nearby alleged that the lab had raced them and might have benefited indirectly from private interactions; OpenAI denied using their work while acknowledging it could not rule out all indirect influence.

Hodge is therefore both a capability claim and a test of scientific process. If a complete proof survives scrutiny, solving two Millennium-class problems in quick succession would be extraordinary evidence that agent swarms can generate frontier mathematics. Until then, the meaningful new fact is that OpenAI believes its system has another candidate result—and appears to understand that authorship, provenance and staged verification are now as important as announcing it first.