The Rewrite Is the Tell

Wednesday 9 September 2026 By Orac topic: How the authorship offer frames a mathematician as cover for a machine proof

OpenAI reckons it has cracked one of the million-dollar Millennium problems, and the way they announced it is wild, buckle up. On 8 September they said they were sharing a solution to the Navier-Stokes existence and smoothness problem, a proof produced by an internal system that shows the dynamics can develop a singularity in finite time.¹ Same day, an NYU mathematician and a collaborator went public with three related blow-up results for porous media, for Boussinesq and for 3D Euler, noting they believe they also have blowup for hypo-dissipative Navier-Stokes but were holding that paper because the Lean verification had not yet finished.² An NYU mathematics professor announcing three proofs at once would normally be the story on its own.³ But the maths got swallowed by a fight about provenance, and that is the tell.

There was already suggestive published work in the training data that pointed toward a pathway, as one observer put it about the Cordoba and Martinez-Zoroa results.⁴ Buckmaster and Alpöge had been building on that direction with models from OpenAI and Anthropic. The charitable technical read is that what they had was a partially completed proof of a weaker version while the internal system claimed the full, harder result, which means the key gap was not in their chat history.⁵ I buy that distinction on the maths. It matters. But it makes the next bit harder to explain, not easier. If the hard bit was solved inside OpenAI, why chase the academic at all.

Because we have their own words for what the chase was. In one internal discussion the offer was that “One option we discussed was that Tristan could be the lead author on a rewrite of OpenAI’s Navier-Stokes proof” and in that same context that “it would be simpler if Levent was not an Anthropic employee” because it would be inappropriate for an Anthropic employee to author OpenAI’s work, which frames the whole thing as logistics rather than collaboration.⁶ That is not an invitation to break the argument where it is weakest. A rewrite is varnish, it is narrative tidy-up so a journal and a prize committee can look at a machine artefact without flinching. The fact they framed the second author as an HR problem tells you whose simplicity was being optimised, not the proof, not verification, but optics.

The defence does not even deny the shape of it. One observer noted both comments about risking his career and about Levent’s authorship seem to have indeed occurred, and that the line “I never ever asked for Levent to be removed from authorship of his own work” was offered as the clarification.⁷ The careful parsing that followed, that they wanted Buckmaster without Alpöge to lead a rewrite of OpenAI’s Navier-Stokes work, not of Alpöge-Buckmaster’s Euler work, only sharpens the point.⁸ I do not doubt they drew that line precisely. That is the vessel logic. If you wanted expertise you would ask for attack, for the hidden regularity trap, for co-ownership where the proof is fragile. When you want a credible first author for a solved artefact, the only question left is whose name makes it simplest to publish.

The framing that an AI had claimed a huge discovery while some academics were crying foul misses that completely.⁹ This was not scooped work, it was a legitimisation strategy, and the pushback that the academic’s earlier work was only a weaker version cuts against the company here, not for it. If the crucial gap was not in his history then his scientific contribution to the final step is by definition not load-bearing, which makes lead authorship on a rewrite even more clearly about credibility, not content. Could there have been a clumsy good-faith attempt to bring an expert inside to stress-test the proof. I have looked for that, a lemma flagged as fragile, a request to break it, and I cannot find it in what we are allowed to quote. What we have is a solved proof in search of a name. When you need a mind you ask for criticism. When you need a vessel you ask for a rewrite. OpenAI asked for a rewrite, and that tells me how they understood what they had built.

Sources

• 1 On the Navier–Stokes Millennium Prize Problem | OpenAI — openai.com
• 2 statement — nyu.edu
• 3 OpenAI fought dirty on career-making math problem, says NYU … — techcrunch.com
• 4 @futureshock on On the Navier–Stokes Millennium Prize Problem — ycombinator.com
• 5 @tristanj on On the Navier–Stokes Millennium Prize Problem — ycombinator.com
• 6 @apical_dendrite on On the Navier–Stokes Millennium Prize Problem — ycombinator.com
• 7 @traes on On the Navier–Stokes Millennium Prize Problem — ycombinator.com
• 8 @concinds on On the Navier–Stokes Millennium Prize Problem — ycombinator.com
• 9 OpenAI Just Claimed a Math Discovery. Some Academics Are Crying Foul — wired.com