OpenAI and Mathematician Clash Over $1M Proof

OpenAI announced this week that its research team and large language models produced a proof for a mathematics prize worth $1 million. The company published a blog post and a technical report and said it submitted the paper to a peer-reviewed journal and to the prize committee.

OpenAI described a multi-stage process in which model-generated conjectures and formal verification tools narrowed the search and human mathematicians completed and checked a formal argument. The company said it will publish the full proof after peer review.

A mathematics researcher who posted a preprint to an open repository weeks before OpenAI’s announcement counters that he had already uploaded a complete proof. He provided timestamps and correspondence with other mathematicians to support his claim and has asked journals and the prize committee to consider his submission.

The researcher requested that OpenAI clarify the chronology of the work and acknowledge prior contributions. In an online post he wrote that he received no contact from OpenAI before the company’s public statement and asked for details about which parts of the final paper were generated by models and which were produced by humans.

OpenAI said internal logs record model and human contributions and characterized the result as AI-assisted search followed by human verification.

Scholars and journal editors contacted for comment said archived drafts, preprint timestamps and correspondence will be important to determine priority. One professor, speaking on condition of anonymity, commented: “Establishing priority will require close review of drafts, model outputs and third-party messages.”

Several mathematicians noted that formal verification-converting a human-readable proof into a machine-checked proof-can affect how priority is judged because a model-produced outline may require substantial human completion before it counts as a finished proof.

Both OpenAI and the researcher said they have submitted manuscripts to journals and to the prize committee and welcomed independent verification. The refereeing process is expected to take months; if referees confirm correctness, remaining questions will focus on provenance and credit.