Tristan Buckmaster accuses OpenAI of stealing math research

Following announcements earlier in September that OpenAI's agents solved the Navier-Stokes physics problem, the narrative has taken a sharp turn. NYU mathematician Tristan Buckmaster accused OpenAI of using prompt data to front-run his proprietary solution.

For months, Buckmaster worked alongside Anthropic researcher Levant Alpoge on a proof for the Millennium Prize equation. Throughout the project, the pair ran their draft code through OpenAI’s Codex tool to assist their calculations.

As first reported on This Week in Startups, OpenAI launched a targeted internal project in September only after hearing rumors about Buckmaster’s progress. The company deployed massive compute resources to replicate the pair's specific, unconventional mathematical direction.

OpenAI executive Sebastian Bubeck rejected claims of theft. However, OpenAI's internal write-up confirmed the project began only after hearing industry rumors. Buckmaster further stated that OpenAI executives pressured him to remove Alpoge from the research paper due to corporate rivalry with Anthropic.

The conflict intensified on Breaking Points, where details emerged that OpenAI could not rule out training its models on Buckmaster's private Codex sessions. When Buckmaster threatened to go public, an OpenAI representative explicitly warned him against ruining his academic career.

On The Intelligence from The Economist, science correspondent Sam Weigley explained that OpenAI deployed 100 autonomous agents running on GPT-6 Astra, spending over $1 million in compute to rush their paper out ahead of the human mathematicians.

The published paper lacks the rigorous structural proofs standard mathematical peer review requires. OpenAI waived the $1 million prize while failing to rule out prompt contamination, fueling accusations that private user inputs were scraped to claim unearned credit.