Tristan Buckmaster accuses OpenAI of stealing math solution

OpenAI's claim to have solved a famous Millennium Prize mathematics mystery is unravelling under accusations of corporate plagiarism.

For months, NYU mathematician Tristan Buckmaster and Anthropic researcher Levant Alpoge used OpenAI's Codex tool to draft proofs for the 90-year-old Navier-Stokes problem. Buckmaster alleges that after feeding draft code into the commercial interface, OpenAI leveraged those private prompts to launch a targeted internal project in September.

The story evolved rapidly across technical channels. On September 8, 2026, OpenAI researchers Metab Swani and Mark Selke outlined on The a16z Show how frontier models solve open problems by backtracking and pruning search paths. But by September 11, The Intelligence revealed the scale behind OpenAI's Navier-Stokes announcement: an autonomous agent swarm running on GPT-6 Astra, burning through over $1 million in compute to draft a solution in under four days.

"The models backtrack when stuck, update internal confidence, and prune dead ends."

Metab Swani, The a16z Show

The speed came at the expense of mathematical rigor.

Science correspondent Sam Weigley reported on The Intelligence that OpenAI rushed its agent swarm into production after hearing rumors that human mathematicians were closing in on a proof. Traditional peer review demands structural insight, yet OpenAI's paper lacked detailed proofs. When Buckmaster questioned whether private session data trained the model, OpenAI denied inspecting user records directly but refused to rule out training contamination.

"OpenAI rushed its model into production after hearing rumors that human mathematicians Tristan Buckmaster and Levant Alpoge were closing in on a solution."

Sam Weigley, The Intelligence

The corporate response escalated quickly. On This Week in Startups and Breaking Points, details emerged of executive pushback. When Buckmaster threatened to go public, an OpenAI representative explicitly warned him against ruining his academic career. OpenAI executive Sebastian Bubeck rejected the theft framing, though the company's official write-up acknowledged starting its effort only after hearing industry rumors. To stem criticism, OpenAI waived the $1 million Millennium Prize.

Industry observers view the episode as a stark warning for academic science. On This Week in Startups, Canvas Ventures co-founder Rebecca Lynn noted that tech giants historically build on external data, cautioning researchers against feeding proprietary work into commercial LLMs. Agent Fund general partner Yohei Nakajima added that autonomous web-scraping agents now aggressively index preprints, stripping independent scholars of traditional research secrecy.

Feeding proprietary breakthroughs into commercial models has become an existential gamble for working scientists.