OpenAI fluid-proofs credit dispute (Buckmaster allegations vs Bubeck / OpenAI replies)
Vintage: 2026-09. Contested. Morning primary allegation: Buckmaster’s statement PDF. Morning denial: @SebastienBubeck. Afternoon issuer + X replies in 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5. Keep parallel attributed accounts. Do not flatten to “plagiarism proven” or “allegations false.”
OpenAI fluid-proofs credit dispute (Buckmaster allegations vs Bubeck / OpenAI replies)
One-line summary: On 8 Sep 2026 tristan-buckmaster alleged that sebastien-bubeck pressed authorship and announcement deals that would sideline levent-alpoge; morning Bubeck called the allegations “false and inflammatory.” Afternoon OpenAI/Bubeck/Altman posts give a concurrent-work / coordination narrative. Keep both sides. The math of OpenAI’s forced-NS C/D claim lives on openai-forced-navier-stokes-claim — do not use this page as proof that the theorem is true or stolen.
The insight
This is a lab-credit conflict, not a resolved fact pattern. Buckmaster’s PDF includes explicit non-claims: he has not seen OpenAI’s proof; does not know what the model did; is not accusing misuse of data — stating what he was told. Bubeck’s X is a general denial plus “more tomorrow.” Discourse that treats Codex-log theft as proven is an overclaim relative to those non-claims.
The Buckmaster–Alpöge math announcement lives on buckmaster-alpoge-forced-blowups. OpenAI’s afternoon issuer NS claim lives on openai-forced-navier-stokes-claim. Do not use this page as evidence that OpenAI solved (forced or unforced) Navier–Stokes, or that either side’s credit story is settled.
Evidence
All bullets are from 2026-09-08-x-morning-buckmaster-alpoge-ai-fluid-proofs-openai-credit. Permalinks live on the source page.
Buckmaster allegations (statement PDF)
- From 2026-09-08-x-morning-buckmaster-alpoge-ai-fluid-proofs-openai-credit (statement PDF): after Sep 3 rumors that Anthropic had resolved a major open problem, Buckmaster emailed an OpenAI-affiliated mathematician; later Sep 6 calls with sebastien-bubeck (levent-alpoge not on the calls) claimed an internal OpenAI model produced a ~100-page forced Navier–Stokes blowup (“options c and d in Fefferman,” smooth forcing).
- From the same source (same PDF): Buckmaster says he was first told “very little human input,” then that a team / large compute / Codex-written prompt were involved; first prompt timing “in the past few days” after word of their work reached OpenAI. He asked whether the model was trained on / accessed their Codex session drafts; says he was told the model does not look up user data, then got no answer on training.
- From the same source (same PDF): two proposals alleged — coordinated announce (their Euler, OpenAI NS next day), or Buckmaster alone writing OpenAI’s NS result while dropping Alpöge (twice, citing Anthropic employment as “annoying”). He declined; alleges “Why would you ruin your career?” / “If you don’t want me to be nice…” response.
- From the same source (same PDF): explicit non-claims — has not seen OpenAI’s proof; does not know what the model did; is not accusing misuse of data.
Bubeck denial + OpenAI-side color
- From the same source (@SebastienBubeck, 2026-09-08): “A series of false and inflammatory allegations against me are currently circulating… I came into the discussion following academic norms… academic standards are of the highest importance to me. Will have more to say tomorrow.”
- From the same source (@polynoamial): Noam Brown quote-posts Bubeck — “really sweet guy with great intentions,” “week long collaboration… worked out,” “sad that we didn’t get to finish it in the way we wanted. More to say tomorrow.”
Afternoon OpenAI / Bubeck / Altman replies (2026-09-08 ~1:20–2:40pm ET)
All bullets in this subsection are from 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5. Parallel accounts — not a reconciliation of the morning PDF.
- From 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5 (issuer Concurrent work section): effort began Sep 1 after a rumor later tied to levent-alpoge (Anthropic) and tristan-buckmaster; after Lean verify (~Sep 6) OpenAI reached out believing they also had NS; learned they had forced Euler; offered concurrent release / priority recognition; congratulate forced-Euler priority; deny seeing their work pre-public release; “no specific user data was accessed”; cannot rule out de-identified product-usage data helping models; proofs differ (Euler forced vs OpenAI unforced).
- From the same source (@OpenAI, 2026-09-08 17:23:27Z): congratulate Alpöge & Buckmaster; same no-user-data / forced-vs-unforced Euler language as the blog.
- From the same source (@SebastienBubeck, 2026-09-08 17:39:53Z): long clarification — screenshot of outreach to Levent; denies asking to remove Levent from his own Euler authorship; says discussion was about Tristan as lead on a rewrite of OpenAI’s NS proof and Anthropic-employee IP awkwardness; apologizes for “risk their career” wording; alleges threats of press accusations; Levent declined meetings. (photo)
- From the same source (@SebastienBubeck): technical points — started after viral rumors Anthropic resolved Millenniums; did not see their work until public; team lacked fluid-dynamics research expertise (multi-agent attempt); Levent’s public posts read as Anthropic-linked.
- From the same source (@sama): defends Seb/team integrity; wanted joint release; when learned Euler≠NS offered them go-first / Tristan lead rewrite of OpenAI proof; felt hard to offer same to Anthropic employee who wouldn’t talk; preferred coordination; approaches appear different now that public.
Secondary / discourse (not grain)
- From the same source: OfficeChai on Bubeck denial, OfficeChai on Buckmaster allegations — secondary press retell.
- From the same source (@hosseeb, @ns123abc): discourse that flattens “OpenAI stole Navier–Stokes from Codex logs” as proven fact — overclaims; flag, don’t promote as grain.
What this source does not establish
- Morning: OpenAI NS writeup was unfetched and unseen by Buckmaster (his own non-claim). Afternoon: OpenAI published an issuer NS page + PDFs — that is a new object, filed on openai-forced-navier-stokes-claim, not a verdict on the credit fight.
- Codex-training / session-access question still unanswered as a settled fact. Afternoon OpenAI: no specific user data accessed; de-identified product-usage data not ruled out.
- Not a Clay prize win and not independent verification of the Lean repo.
- No Noam Brown person page. Secondary morning color only.
- No ticker.
Contradictions / tensions
- Allegation vs denial vs afternoon clarification. Buckmaster’s PDF vs morning Bubeck “false and inflammatory” vs afternoon Bubeck (deny drop-from-Euler; rewrite-of-OpenAI-NS / IP-awkwardness / apology for career wording) vs Altman coordination narrative. Not reconciled. Keep parallel attributed accounts.
- Forced Euler vs forced NS vs unforced Euler. Morning public Buckmaster–Alpöge object is forced Euler/Boussinesq/IPM; afternoon OpenAI object is forced NS C/D plus a separate unforced Euler. Do not flatten “they had the same result.”
- User-data. OpenAI denies specific user-data access and leaves de-identified training-data possibility open.
Open questions
- Does any third-party reconstruction match or contradict either timeline (Sep 1 start / Sep 6 outreach / first-prompt timing)?
- Independent review of the two public proof objects — still open on the math pages.