Drama swirls around OpenAI’s legendary mathematical milestone

Details

OpenAI announced it has solved the Navier-Stokes equations, one of seven Millennium Prize Problems each worth $1M, using an undisclosed internal model described as more capable than GPT-6 Astra, deployed with 10,000 concurrent agents. OpenAI began training this model on August 28th, saying it shows 'unprecedented performance' on mathematics benchmarks. The proof would represent the first time AI has cracked a Millennium Prize Problem. The announcement arrived one day after NYU mathematics professor Tristan Buckmaster and Levent Alpöge — a researcher at Anthropic — published findings on a closely related problem. Buckmaster says OpenAI's proof follows the same mathematical route he and Alpöge had been developing, using OpenAI's own Codex tool where they stored all their working drafts throughout the project. Buckmaster asked OpenAI directly whether the model had been trained on their Codex sessions. He received a denial about direct data access but no clear answer on training. OpenAI's official statement says 'no specific user data was accessed' but acknowledges it 'cannot rule out that de-identified data derived from their usage helped improve our models' — a significant hedge that leaves the ethical question unresolved. The dispute exposes a structural vulnerability: researchers using commercial AI tools to develop unpublished work may inadvertently feed that work into model training. OpenAI senior researcher Sebastien Bubeck maintains the proofs differ significantly, but the overlapping mathematical route remains publicly unexplained.