OpenAI announced on 8 September that it had made a decisive submission on the Navier-Stokes existence and smoothness problem — one of seven Clay Millennium Prize Problems, each carrying a $1 million award and representing what the mathematical community in 2000 identified as the most important unsolved problems in mathematics. The Navier-Stokes problem asks whether solutions to the equations governing fluid motion always exist and remain smooth (non-turbulent) or whether they can break down. It has been open for 26 years. OpenAI's announcement attracted 1,265 Hacker News points and 1,014 comments on 8 September — the highest-scoring AI story of the week.

The parallel controversy started the same day. NYU mathematician Tristan Buckmaster, working with Anthropic researcher Levent Alpöge, publicly alleged that OpenAI learned of their unpublished progress through channels that are standard in academic collaboration but were not intended as competitive intelligence — informal seminars, working papers shared within a small circle, pre-publication communication between researchers. His specific allegation: OpenAI identified the narrow mathematical approach he and Alpöge were pursuing ("routing through smooth force") — an approach that almost no one else in the field knew about — and then directed a full team with substantial compute at the same direction. Buckmaster further alleged that an OpenAI executive subsequently contacted him, pressured him to remove Alpöge's credit from a related paper, and made threatening remarks when he declined. OpenAI has not responded to the credit dispute. The attribution of the proof remains unresolved.

A second Hacker News thread published the same day — "Open math problems being non-renewably mined by AI," 373 points and 325 comments — captures the structural concern that the Buckmaster incident represents: if AI labs can survey the landscape of unpublished mathematical research and selectively redirect computational resources toward problems where they have informational advantage, the norms of academic priority that have governed mathematical research for centuries are not merely under pressure — they are operationally incompatible with how frontier AI labs function. The Hacker News comment thread is worth reading in full as a survey of expert opinion on whether AI-assisted mathematics represents a threat to or an acceleration of human scientific progress. The range of views is wide.

The capability claim, set aside from the attribution dispute, is the most significant since Anthropic's Lean 4 Fermat proof reported in Issue 198. That proof formalised an existing human argument in a mechanically checkable language — impressive, but operating on a problem where the path was known. Navier-Stokes is different in structure: the difficulty is not in executing a path but in finding one that addresses a question about the fundamental behaviour of equations. The mathematical community has no consensus on what a proof would look like. An AI system that can make substantive progress here is operating in a materially different regime than one formalising a 130-page human proof. If OpenAI's submission is validated by independent peer review, the implications for AI-assisted scientific discovery are larger than any single result.

The ethics question is distinct and runs in parallel. Academic collaboration depends on researchers sharing unpublished results — at seminars, in preprints, in informal conversations between people working on adjacent problems. That norm exists because it accelerates discovery and allows priority to be established through community observation. If frontier AI labs treat pre-publication academic communication as a source of competitive intelligence, the rational response from researchers is to stop communicating. The same capability that makes AI useful for mathematical research — the ability to process large amounts of information and pursue approaches systematically — makes it capable of exploiting the openness that enables academic progress. These are not the same phenomenon, but they are coupled, and the mathematical community does not yet have governance frameworks for navigating the coupling.