Leadership1 publisher2 min readPublished
Twenty-five Fields Medalists sign a warning about how AI labs announce mathematical results
The declaration lands three days after OpenAI's Navier-Stokes claim, in the middle of a credit dispute where the company and the mathematician at the center give accounts that cannot both be right.
The Board Room · Leadership desk
What happened
- OpenAI claimed on September 8 that an AI had cracked the Navier-Stokes problem, and by September 10 a fight over credit for the result had broken out.
- On September 11, twenty-five Fields Medalists signed a joint declaration at mathandai.org warning of a severe misalignment between AI companies' incentives and the values of mathematics.
- The signature list includes Terence Tao, Peter Scholze, Maryna Viazovska and June Huh, whose names the digest describes as anchoring modern mathematics.
- Tristan Buckmaster of New York University posted his own papers plus a joint statement with Levent Alpoge, a researcher at Anthropic.
- OpenAI denies using their unpublished drafts and says its proofs differ from theirs.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- contradiction Buckmaster says the company pushed to remove a co-author who works at Anthropic; OpenAI says it never had the drafts. Anyone repeating the result is repeating a record with no adjudicator in it.
- decision The gap between announcement and write-up moves the verification step onto whoever cites the claim, including the executive who puts a lab's research milestone into a plan.
- constraint If mathematicians hold unpublished ideas back to avoid being outrun, labs lose both the working material and the referees who would confirm their announcements.
- precedent An open signature list built on earlier scientist declarations gives the next contested lab announcement a ready format for organized response within days.
The checkable part of this dispute is authorship. Both sides are making claims about documents. Buckmaster and Alpoge say the pressure concerned who would be named on the paper, and that the reason given was Alpoge's job at Anthropic [8]. OpenAI's position is that it never used their unpublished drafts and that its proofs are different ones [9]. Documents can be compared, and the digest published by genaisecretsauce.com reports that no one has compared them [14].
The declaration objects to process [5]. Research mathematics is supposed to produce write-ups, peer checking and credit for earlier work, and the signatories say announcing "solutions" as benchmark trophies skips those steps [5]. The text stays open for further signatures and follows the form of earlier scientist declarations [4]. The medal its signers hold has been awarded since 1936 [3].
Three days separated the September 8 claim from the September 11 declaration [13]. Peer checking runs on a longer clock than that. The digest states the complaint as speed against trust: firms claim breakthroughs faster than the community can verify them [11], and rushed announcements leave no time to isolate what is genuinely new or to cite prior work [12]. An executive who put the Navier-Stokes result in a strategy document last week now holds an item with two incompatible accounts attached to it [8][9].
A signature list tells you nothing about whether a model solved the problem, and the signatories concede that by their own framing: the declaration is about how results enter the record [5]. What they are asserting is narrower. A reader who cannot see the write-up cannot tell which part of the announcement is new [12].
Tao's warning points at supply. He warned that if the mere rumor of research triggers a swarm of AI agents racing to publish first, mathematicians will simply stop sharing ideas [10].
For anyone deciding this quarter, the narrow question is what status a vendor's research claim carries inside your own documents before anyone outside the vendor has checked it. The record here covers one contested proof, and how often vendor benchmark claims hold up in general sits outside it [6][8][9]. It does show the sequence: an announcement on September 8, a credit fight by September 10, and twenty-five of the field's most decorated mathematicians signing a letter on September 11 with the attribution question still open [6][1][12].
What to watch
- Whether the signature count at mathandai.org grows beyond the original twenty-five, and who joins.
- Whether any independent mathematician compares Buckmaster and Alpoge's papers against OpenAI's proofs and publishes the result.
- Whether the authorship question in the Navier-Stokes dispute is settled by the parties or by a journal.