OpenAI used work from mathematicians to solve Navier-Stokes problem, then attempted to cover its tracks
OpenAI allegedly offered compute in exchange for information.
Sam Altman, OpenAI. By REUTERS/Dado Ruvic
- OpenAI says Astra, its GPT-6 model, solved the 90-year-old Navier-Stokes problem.
- Tristan Buckmaster and Levent Alpöge say they had already developed and announced related results.
- Buckmaster alleges OpenAI used their approach, overstated its model’s role, and offered free compute.
- The dispute centers on credit for a possible breakthrough in a long-standing fluid motion problem.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
OpenAI allegedly copied a pair of mathematicians' approach to solving the Navier-Stokes problem, embellished the truth, and attempted to butter them up with free compute.
OpenAI claims its newest AI model, Astra (GPT-6), has successfully solved the 90-year-old Navier-Stokes math problem.
The Navier-Stokes equations were first developed by Claude-Louis Navier and George Gabriel Stokes in the 19th century to further our understanding of fluid flow, such as that of water and air.
In a blog post, the AI lab has provided proof “produced by an internal OpenAI system,” showing that the dynamics of the equation for fluid motion can develop “a singularity in finite time.”
Not only is this a huge discovery in the mathematics community, but it also demonstrates “the sheer magnitude of what frontier models can now do.”
At least that’s what Tristan Buckmaster, one of the mathematicians who allegedly solved the Navier-Stokes equation first, said.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
Mathematicians used OpenAI’s tech to solve Navier-Stokes
Buckmaster and fellow mathematician Levent Alpöge announced 3 public results, including the “finite-time blowup with smooth forcing for incompressible porous media.”
Google shows that Buckmaster’s statement was published 4 days ago, on September 6th, 2026.
The pair embarked on a personal venture to solve the almost century-long problem without any “institutional agreement or official involvement by either of our employees,” Buckmaster said.
Alpöge works as technical staff at Anthropic, the maker of Claude, and Buckmaster works as a mathematics professor for the Courant Institute of Mathematical Sciences at New York University.
Using several large language models like Anthropic’s Claude, OpenAI’s Codex (specifically GPT-5.6 Sol), and most recently GPT-6, OpenAI’s latest model, Astra, the pair pushed and iterated the basic idea, which was developed by 2 different mathematicians.
“The credit for the basic idea of this program goes to Diego Córdoba and Luis Martínez-Zoroa, who for several years have been exploring the construction of forced blow-ups,” Buckmaster said.
While this discovery should’ve been celebrated, the mathematicians were forced to get their findings out as quickly as possible, with Buckmaster expressing his dissatisfaction with the “presentation quality in these papers.”
But the mathematicians needed to get ahead of OpenAI’s alleged plans to eclipse their announcement.
OpenAI allegedly lied, cheated, and forced the problem
Around September 3rd, 2026, Alpöge received information that the pair's progress had been passed on to OpenAI, following the news that Anthropic had solved a major open problem.
Buckmaster emailed a prominent OpenAI mathematician, saying that the rumor mill was hard at work, with many people from various institutions discussing a potential solution traced back to Alpöge's work, and that their research was self-funded.
He also noted that the pair had put money directly into OpenAI’s pocket during the process.
The OpenAI mathematician responded that day, asking Buckmaster for details that would be “useful to avoid competing.”
They also said, “We are always thrilled when mathematicians make progress with our models,” and if there’s “anything in terms of compute from OpenAI’s end, we would be happy to provide it.”
Following discussions with only Buckmaster, in which an OpenAI technical staff member, Sebastian Bubeck, was allegedly present, the mathematician was told that an “OpenAI model had produced a proof of finite time blowup for the forced Navier-Stokes equations.”
When asked in writing, Alpöge was told that OpenAI had chosen to approach the problem in the same way that Buckmaster and Alpöge had.
“The route to the Clay…is the route Luis and Diego opened, and the one Levent (Alpöge) and I had quietly chosen to attack.”
Alpöge had also allegedly been told by Bubeck that “very little human input” had been used to crack the problem.
“This turned out not to be true,” Buckmaster said.
