OpenAI claims Millennium Prize breakthrough, but secrecy sparks academic backlash

OpenAI announced Tuesday that it has solved one of mathematics' most celebrated open problems, a Millennium Prize question that has stumped researchers for decades. The result, if verified, would be a landmark achievement for artificial intelligence and a stark demonstration of how quickly machines are reshaping mathematical research. But the triumph was complicated almost immediately by the unusual way the company chose to share—and not share—its work.

The announcement was met with a mix of awe and unease across the academic community. While researchers acknowledged the technical depth of the solution, many were frustrated by the lack of full methodological transparency. OpenAI has not released the complete proof or the underlying reasoning process, citing competitive and safety concerns. That has left mathematicians unable to verify the result independently, a step long considered essential in the field.

The Millennium Prize problems, a set of seven questions posed by the Clay Mathematics Institute in 2000, each carry a $1 million reward for a correct solution. Only one has been solved to date—the Poincaré conjecture, proved by Grigori Perelman in 2003, who declined the prize. OpenAI's claim, if confirmed, would mark the second resolution and the first attributed to an AI system, raising immediate questions about how credit and prize money should be allocated.

The episode highlights a growing tension between the open norms of academic mathematics and the proprietary practices of leading AI labs. Over the past year, AI tools have begun suggesting conjectures, checking proofs, and even finding new connections between distant areas of math. But most of that work has been published openly, with code and data shared for scrutiny. OpenAI's decision to hold back key details marks a departure that some researchers see as a threat to the integrity of the field.

Efforts to verify the result are now underway at several university departments, though progress is slow without access to the full proof. Some mathematicians have called on OpenAI to release the work to an independent review board, while others argue that the company's commercial interests should not dictate the pace of scientific validation. The Clay Mathematics Institute has not yet commented on whether it will consider the claim.

For now, the breakthrough stands as both a technical marvel and a cautionary tale. It shows what AI can achieve when pointed at problems that have resisted human effort for decades. But it also raises a hard question: if the next great mathematical discoveries come from black-box systems that refuse to show their work, can the wider community still trust the results? That answer may determine whether AI becomes a collaborator in mathematics or a source of new division.