Science

Mathematician alleges OpenAI took his work in Navier-Stokes proof claim

Scientific American reports that OpenAI claimed to solve Navier-Stokes amid a public dispute over credit and method.

Updated Oct 22, 2026 · 18:15 UTC1 reports

The short version

Scientific American reports that OpenAI claimed on September 8 to have solved one of math's million-dollar problems.[Scientific American]
The report says OpenAI's internal model reached a correct answer by a path its top mathematicians initially did not anticipate.[Scientific American]
Tristan Buckmaster publicly accused OpenAI of seeming to take his work, and the report says OpenAI's claim of a shared method proved wrong.[Scientific American]
Unresolved: Bubeck reportedly first said the model solved the problem in one go, then described a weeklong agent effort.[Scientific American]

On September 8, OpenAI claimed to have solved one of math's million-dollar problems, shortly after Tristan Buckmaster publicly accused it of seeming to take his work.[Scientific American]
On August 22, Buckmaster and Alpöge prompted their language models to code their proof in Lean, a language that breaks a proof into small steps and checks each one.[Scientific American]
According to the report, OpenAI began training a new internal model on August 28 that proved far better at math than any predecessor.[Scientific American]
On August 31, Alpöge posted a Latin phrase meaning miraculous August, which OpenAI's team reportedly read as a sign that Anthropic had solved at least one Millennium Problem.[Scientific American]
A squad of 100 OpenAI agents reportedly removed the forcing term from the Euler approach and produced its own Euler proof, with the fluid blowing up without being stirred.[Scientific American]
A 10,000-agent swarm reportedly solved Navier-Stokes after consuming a reported $15 million in computing resources, and had to reintroduce a force to do so.[Scientific American]
Bubeck reportedly first claimed the model solved the problem in a single go from the official statement, then described the weeklong agent effort.[Scientific American]
Bubeck offered Buckmaster lead authorship on OpenAI's Navier-Stokes paper, on condition that Alpöge's name be removed because of his affiliation with a competitor.[Scientific American]

Credit is contested: Buckmaster alleges OpenAI relied on his work, while the report says OpenAI's statement of a shared proof path proved wrong.[Scientific American]
Bubeck's account reportedly changed during a call, so the sequence and timing of OpenAI's proof work remain unclear.[Scientific American]

Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.