Researchers question trusting OpenAI with unpublished math
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
OpenAI says its AI solved the Navier-Stokes Millennium problem; mathematicians ask if their unpublished work helped.
The dispute is about whether private research fed into models that then compete with.
Background
OpenAI announced that an unreleased multi-agent model solved the Navier-Stokes Millennium Prize Problem, a roughly 90-year-old question about whether smooth 3D fluid motion can break down. Mathematicians Tristan Buckmaster and Levent Alpöge had put related unpublished drafts into private Codex sessions, and Buckmaster asked whether that data was used or accessible. OpenAI says no specific user data was accessed, but acknowledges it cannot rule out that de-identified data derived from their product usage helped improve its models.
How it unfolded
- Sep 1, 2026OpenAI says its researchers heard rumors that mathematicians at rival Anthropic were close to solving two Millennium problems, and deployed an in-development model to try to crack one first.
- Sep 8, 2026OpenAI announces its model solved the Navier-Stokes problem; reports say it took days and millions of dollars in computing power.
- Sep 9, 2026Mathematician Andreas Thom posts that the episode exposes an unanswered question about whether researchers can trust OpenAI with unpublished mathematics, citing his earlier exchange with OpenAI over the non-sofic-group announcement.
- Sep 10, 2026Coverage continues as mathematicians raise concerns over timing, unpublished work and credit; OpenAI denies a claim that it offered to collaborate if Alpöge's name was removed.
Who’s saying what
- OpenAI
- OpenAI says its researchers and AI agents did not see the pair's work before it became public and that no specific user data was accessed to solve the problem, while acknowledging it cannot rule out that de-identified data from their product usage helped improve its models.
- Researchers
- Andreas Thom says the episode exposes a serious, unanswered question about whether researchers can trust OpenAI with unpublished mathematics, and that the lack of transparency could damage the communal process of math.
- Analysts
- Commentators say the provenance question remains unresolved on publicly available evidence, and that the harder issue is what a platform is allowed to do with what gets typed into it and how influence could ever be proven after the fact.
Still unverified
Buckmaster's account, as characterized in commentary, suggests OpenAI may have used researchers' unpublished work to race them to a breakthrough and questions the attempt to manage scientific credit; those posts are commentary, not independent evidence that OpenAI stole a proof. Buckmaster says he asked whether their data had been used and received no answer, and that OpenAI offered to collaborate if Alpöge's name was removed because of his Anthropic affiliation; OpenAI denies this claim. Whether OpenAI's proof is correct has not been decided by mathematicians.