FULL STORY

OpenAI Accused of Scooping Navier-Stokes Breakthrough

Mathematician Tristan Buckmaster accused OpenAI of rushing out their LLM-assisted Navier-Stokes work and dropping his name; Sam Bubeck denied the claims, sparking ongoing dispute.

2026-09-08 ~ 2026-09-08 · 3 episodes · 29 posts

Episode 1 · OpenAI Accused of Scooping Navier-Stokes Work; Bubeck Denies Allegations (2026-09-08, 19 posts)

The controversy stems from a dispute between OpenAI and independent researchers including Tristan over attribution of Navier-Stokes equation solving research: reports claim OpenAI may have accessed private Codex session data from Tristan and Levent, based on OpenAI's refusal to confirm it did not train on that data.

Confirmed

  • The core allegation, relayed by @Jsevillamol, is that OpenAI declined to confirm it did not use the two mathematicians' private Codex sessions for training, which was read as possible access to that data.
  • OpenAI researcher willdepue publicly responded that such snooping could not happen, saying OpenAI, like other large regulated platforms, has strict controls on user data access.

Unconfirmed

  • Whether OpenAI actually accessed or used private Codex session data remains an indirect inference (a refusal to deny), with no direct evidence.
  • @generatormanai noted community debate over research priority: one side argues Tristan sharing his work with OpenAI steered the model toward that research direction, unlike parallel discovery; the other side points out there is no evidence Tristan ever made such a share.

Why it matters

  • @basedjensen, from an industry insider's perspective, argues that "training on private Codex sessions" would require bypassing multiple layers of internal security controls—unlikely unless ordered at the institutional level—and doubts OpenAI leadership would take on that liability risk. This provides a framework for evaluating the rumor but leaves open the possibility of institutional action.
  • @teortaxesTex flagged a deeper concern: if OpenAI can freely read users' Navier-Stokes sessions, it implies routine scanning of user sessions to "snipe" valuable results—shaking the very foundation of user trust in tools like Codex.

Episode 2 · Mathematician Claims LLM-Assisted Breakthrough, Accuses OpenAI of Scooping and Co-Author Removal (2026-09-08, 8 posts)

Tristan Buckmaster, a mathematician at New York University, issued a public statement around September 8, revealing that he and Levent Alpöge, a researcher at Anthropic, spent roughly a year studying variants of the Euler, Boussinesq, and Navier-Stokes equations, heavily using tools such as Claude and Codex to generate candidate ideas, ultimately cracking several major fluid dynamics problems within about a month. The Navier-Stokes problem is one of the seven Millennium Prize Problems, carries a $1 million reward, and is second in fame only to the Riemann Hypothesis and P vs NP.

Confirmed

  • Buckmaster's statement contains two accusations: first, that after OpenAI learned the two were on the verge of publishing their proof first, it attempted to rush out similar results ahead of them; second, that OpenAI demanded his collaborator be removed from the author list on the grounds that he worked at Anthropic. IgorCarron relayed that the episode has been called one of the ugliest incidents ever to come out of a frontier lab — the issue lying not in the mathematics itself, but in OpenAI's conduct after learning someone else was about to beat them to it.
  • According to eyishazyer, after the dispute escalated, Buckmaster accused OpenAI of cutting corners and refused to answer related questions.

Unconfirmed

  • OpenAI's response to the accusations, as well as the actual content and correctness of its "rushed" version of the results, are not provided in the post.
  • The specific mathematical details of both sides' proofs are not presented in the material, and there is no third-party ruling on who was first.

Why It Matters

  • This is a rare, high-profile case of a priority dispute between an AI lab and academia: AI involvement in research is already accelerating mathematical breakthroughs, while issues around lab competition and authorship norms are coming to the fore.
  • Researcher Afinetheorem commented that in an era of AI-accelerated research, "word-of-mouth about an idea" may be enough for someone to crack a problem first, and secretive research may become the norm; he argued that fairness means letting the original authors write up their results at their own pace.

Episode 3 · Researcher Dismisses Claims That OpenAI Steals Ideas From User Chats (2026-09-08, 2 posts)

Researcher BlackHC argues that accusations of OpenAI mining user chats for research ideas are implausible, as the reputational risk of exposure would be catastrophic for the company.