NYU Mathematician Accuses OpenAI of Burning $22.5M in Compute to Hijack Millennium Problem Proof
A mathematician used an AI coding assistant to refine an unreleased proof — days later, a tech giant unleashed tens of millions in compute to beat him to the finish line.
On September 8, 2026, NYU mathematics professor Tristan Buckmaster released three proofs making preliminary progress on the Navier–Stokes existence and smoothness problem — one of the seven Millennium Prize problems carrying a $1 million bounty from the Clay Mathematics Institute. The research was conducted in collaboration with Anthropic mathematician Levent Alpöge, relying heavily on OpenAI’s Codex assistant. Shortly after Buckmaster's statement, OpenAI abruptly published its own complete proof, claiming an unreleased next-generation model had solved the problem.
According to Buckmaster’s public statement, the duo had quietly pursued an obscure theoretical route through smooth forces. While finalizing their results, they learned that information regarding their progress had reached OpenAI. OpenAI began prompting its models on September 1, consuming 300 billion output tokens — worth approximately $22.5 million in compute at current Astra rates — over just one week to rush out a solution. Buckmaster noted that this specific approach is not something a model stumbles upon from a raw prompt, raising concerns that his extensive inputs into Codex informed the model. In its response, OpenAI conceded it 'cannot rule out that de-identified data derived from their usage' helped improve the models.
The private interactions proved even more alarming. Buckmaster alleged that OpenAI mathematician Sébastien Bubeck pressured him to remove Alpöge's co-authorship due to his affiliation with rival lab Anthropic. When Buckmaster insisted on disclosing the dispute publicly, Bubeck reportedly challenged him: 'Why would you ruin your career?' followed by: 'If you don't want me to be nice, then I don't have to be nice.' Following the controversy, Fields Medalist Terence Tao commented that fertile open mathematical problems are now being 'non-renewably mined' by AI.
Sources: TechCrunch reporting by Russell Brandom ('OpenAI fought dirty on career-making math problem, says NYU mathematician') and Professor Tristan Buckmaster's public statement.
BenPig clutched its unfinished draft paper tightly to its chest.