OpenAI Claims Solution to Navier-Stokes Millennium Prize Problem Amid Bitter Data-Ethics Dispute with Mathematicians

OpenAI Claims Solution to Navier-Stokes Millennium Prize Problem Amid Bitter Data-Ethics Dispute with Mathematicians

On Tuesday, September 8, 2026, OpenAI announced a historic technical milestone: an internal AI reasoning model had produced a claimed solution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems carrying a $1 million award each. The 165-page proof and its accompanying formalization in the Lean programming language represent the first time a major unsolved mathematical problem has been resolved by artificial intelligence.

However, the announcement was immediately overshadowed by a bitter dispute. Prominent mathematicians, including New York University professor Tristan Buckmaster and German mathematician Andreas Thom, have accused OpenAI of "dishonesty" and "materially misleading" communications, alleging the company’s models may have covertly benefited from their private ChatGPT and Codex interactions without proper attribution.¹

The Technical Breakthrough: Finite-Time Blowup

The Navier-Stokes equations have governed fluid dynamics since the 19th century, but mathematicians have spent nearly 90 years trying to prove whether smooth, three-dimensional fluid motion can break down under these equations.

OpenAI’s model proved that fluid motion can indeed break down, demonstrating a configuration where a vortex tightens and spins infinitely fast—a phenomenon known as finite-time blowup—while the overall energy of the fluid remains bounded. To achieve the proof, OpenAI deployed up to 10,000 AI agents in parallel over 88 hours, incurring millions of dollars in compute costs. The company stated it has no intention of claiming the $1 million prize administered by the Clay Mathematics Institute.

The Dispute over Private Training Data and "Dishonesty"

The dispute erupted when NYU professor Tristan Buckmaster publicly questioned whether OpenAI's models had siphoned his work. Buckmaster claimed that an OpenAI researcher had pressured him to drop his collaborator from his own related project and warned him that taking the dispute public would damage his career. Sébastien Bubeck, OpenAI's head of mathematics research, later confirmed that the researcher's remarks were "a bad choice of words."

Mathematician Andreas Thom, whose work overlaps with another recent OpenAI mathematical breakthrough (non-sofic groups), also accused the company of "plain dishonesty." Thom began auditing his own ChatGPT interactions and asked OpenAI researchers Sébastien Bubeck and Mark Sellke whether his conversations had been used to train the model.

According to Thom, the researchers' response was narrowly restricted to the question of "direct access" to his chats, failing to clarify whether his conversations had been incorporated into the company's training pipelines as de-identified data.

"De-identification may remove a name; it does not remove the intellectual content of a mathematical idea. ... [Sellke's response was] at minimum, unjustifiably broad and materially misleading; looking back it was plainly dishonest." — Andreas Thom, quoted in The Verge

OpenAI maintained that neither its researchers nor its agents had viewed Buckmaster's or Thom's work prior to public release, but stopped short of categorically excluding that de-identified data generated through their product use had shaped model development.

Strategic Context and the Enterprise Push

The Navier-Stokes solution follows a series of math breakthroughs for OpenAI, including its Astra model generating Lean-verified proofs for 10 longstanding math and computer science problems.

The technical success comes at a critical time for OpenAI. CFO Sarah Friar disclosed that the company is aggressively pushing its models into specialized workflows like chip design, life sciences, and financial services. On September 10, 2026, the company launched "ChatGPT for Financial Services" in partnership with Morgan Stanley and Evercore, utilizing GPT-6 Astra to automate the work of junior investment bankers.

OpenAI is also undercutting Chinese open-source rivals on price, cutting the cost of its Luna model by 80% to make it cheaper than Z.ai’s GLM-5.3. The company hopes these enterprise gains will solidify its commercial standing ahead of a highly anticipated public listing.