OpenAI's Sébastien Bubeck says he reached out to Levent Alpöge to coordinate their releases and denies asking to remove Alpöge from authorship of his own work
and everyone else—acted with integrity and generosity throughout. Initially we believed the other team had also solved the problem. We wanted to collaborate and do a joint release. When we learned that they had Euler but ...Sebastien Bubeck /@sebastienbubeck:Some more technical points: (a) We began working on the Millennium problems due to viral twitter rumors that Anthropic had resolved 2 Millenium problems. Our aim was to see whether our system was also capable of this impressive feat, especia
@sebastienbubeckSebastien Bubeck
Context & Ripple Effects
The dispute followed Tristan Buckmaster’s allegation that OpenAI learned of work he was pursuing with Anthropic’s Levent Alpöge, an allegation OpenAI has contested in a public rebuttal to Buckmaster’s account. OpenAI separately said it could not fully exclude the possibility that de-identified data derived from product use contributed to model improvement, while calling that unlikely.
Bubeck’s account shifts the immediate argument toward release coordination and academic credit. He said work on the Millennium problems began after unconfirmed social-media rumors about Anthropic, while OpenAI has denied that its researchers or models saw Buckmaster and Alpöge’s prompts in its response on prompt access.
First-order effects
Bubeck’s denial makes Alpöge’s authorship and the sequence of contacts a central test of OpenAI’s claim that it sought collaboration rather than appropriation.
OpenAI’s math team must defend its result on both technical independence and credit allocation, rather than treating the work solely as a model-performance announcement.
Second-order effects
The combination of OpenAI’s prompt-access denial and its qualified statement on de-identified product data puts greater weight on provenance records separating researcher access, user interactions, and model development.
Researchers considering AI-assisted work gain a concrete reason to clarify authorship, disclosure, and release coordination before publicizing results that may overlap with labs’ research efforts.
Third-order effects
If high-profile mathematical claims increasingly overlap with external researchers’ work, independent verification and credit protocols may become part of how AI labs establish scientific legitimacy.
The episode points toward accountability for AI research outputs extending beyond whether a system reaches a result to how inputs, collaborators, and attribution are documented.
The trend: AI labs’ pursuit of scientific prestige is making provenance, independent validation, and academic credit as consequential as benchmarked capability claims.
I spent much of the weekend talking with the team who did this work. Seb—and everyone else—acted with integrity and generosity throughout. Initially we believed the other team had also solved the problem. We wanted to collaborate and do a joint release. When we learned that they …
Some more technical points: (a) We began working on the Millennium problems due to viral twitter rumors that Anthropic had resolved 2 Millenium problems. Our aim was to see whether our system was also capable of this impressive feat, especially given our excitement regarding the …
@SebastienBubeck If you're acknowledging that all academic accolades should go to them, how on earth can OpenAI be taking credit for the result? Claiming that it was because of their great new internal model. That is clearly contradictory.
@polynoamial If the system could get web access, then it would have seen the rumours. At that point “Let's look at the logs” is a very natural thing to pursue. And OpenAI has demonstrably got poor situational awareness about this kind of thing
Here is Seb's statement. This is in line with what I sort of expected to have happened after reading Tristan's initial statement One important point of comparison is to compare the results and what was achieved The whole thing is def very messy Congratulations to everyone for the…
seems like OpenAI got totally slandered in all of this Navier-Stokes drama. lots of you were pretty bloodthirsty for it, too. reminder to stop listening to twitter posters whose biases are already established. instead keep an open mind until more evidence comes out
Levent's gimmick is over. Well, we're all in the same boat, just without Anthropic compensation. Even if OpenAI doesn't just release a proof on Github, even if they never announce - you know: it's either solved, or could be solved by them with so much as a wish. Your own problem.
Here is the “other side” of the story. As far as I am concerned, it all seems very reasonable. I don't see any intentional malice and/or underhandedness in how all of this was handled by OAI.
lol it only took them a week for them to solve one of the most important open problems in mathematics... and they only did it because their was a rumor someone else was going to do it first.
I think if you're whatsoever close to math, you're about to publish the solution to Navier-Stokes, and the thing you're worried about is *the appropriateness of Anthropic-OpenAI coauthorship*, you're full of shit and I don't trust anything you say. pox on both your fiefdoms.
I understand how, looking at today's LLMs, people might think the only way we'd achieve NS is by using Levent's/Tristan's prompts. But I hope this plot conveys the model used is a huge step up from today's LLMs. Nobody looked at Levent's/Tristan's prompts. That'd be insane.
Clarifications from Seb. He reached out to Alpöge and Buckmuster under the (mistaken) belief that they have also solved the Navier-Stokes problem, in order to coordinate publication and to avoid scooping them. It turned out they proved a restricted result (forced Euler problem; t…
> It is true that we tried this because there were rumors on the internet One general takeaway from NS results. AI is now fast and powerful enough that mere rumors of an idea working can be enough to crack things.
@SebastienBubeck Thanks for sharing and congratulations on your team's achievement I hope you and Anthropic do manage to get open lines of communication as there will undoubtedly be more breakthroughs to come, it is too awesome a time for it to be otherwise 🙏
when you offered joint authorship of concurrent efforts, you unilaterally tried to dictate who from other side can stay on the paper and who cannot? oh am I seeing this for the first time. (I have had a somewhat related experience; seb probably doesn't remember/don't even know)
An account of recent events by Sebastien Bubeck, which paints quite a different picture. Important for anyone interested in this controversial chain of events to hear from all sides. (Then we can focus on the maths!)
the biggest take away is they solved one of the hardest problems in mathematics, not for humanity nor for the prize money nor for the science, but for the ability to poast about it on twitter faster than the other guy, truly beautiful
Yes, this result cost millions of dollars. But remember that when @OpenAI announced o3 it cost ~$500,000 to score 87.5% on ARC-AGI 1. Today, Astra scores higher for ~$20. In 2025 it took us and GDM an enormous amount of compute to achieve IMO gold. For the 2026 IMO, anyone with a…
OpenAI also acknowledging it did start working on Navier-Stokes after hearing rumors that Anthropic was getting close to a solution. (But implies they did not snoop on their work.) A math fight for the ages! https://x.com/...
OpenAI Just Claimed a Huge Math Discovery. Some Academics Are Crying Foul — A landmark announcement by the frontier AI lab has been overshadowed by accusations of impropriety. WIRED — www.wired.com/story/openai...
Separate from the serious allegations of plagiarism in solving one of the most significant problems in math, you gotta wonder: Would spending millions of dollars on dozens of humans (and vast compute) have *also* worked? — www.wired.com/story/openai...
Today, Levent Alpöge and I have made public three results: finite-time blowup with smooth forcing for incompressible porous media, for Boussinesq, and for 3d incompressible Euler. — https://cims.nyu.edu/... https://cims.nyu.edu/... https://cims.nyu.edu/... https://cims.nyu.e…