OpenAI Claims to Have Solved Navier-Stokes Equations — Authorship Dispute and Technical Rebuttal Erupt on the Same Day
OpenAI's Navier-Stokes proof claim is met with authorship disputes and mathematical rebuttals on the same day.
OpenAI announced that an unreleased AI model produced a complete proof of finite-time blowup for the Navier-Stokes equations, one of the seven Millennium Prize Problems. The announcement immediately triggered an authorship dispute — with an Anthropic researcher allegedly excluded from credit — and a technical rebuttal from Princeton mathematician Stan Palasek, raising questions about whether the proof's core mechanism can survive scrutiny.
An Announcement That Shook the Mathematics World
On September 8, 2026, OpenAI released an announcement that sent shockwaves through the entire mathematics community: an internally unreleased model had allegedly produced a complete proof of finite-time blowup for the Navier-Stokes equations under forced conditions. This is a problem that has confounded mathematicians for over ninety years, and one of the seven "Millennium Prize Problems" listed by the Clay Mathematics Institute in 2000.
Yet the announcement was met not with pure celebration, but with an authorship dispute and technical rebuttals from top mathematicians — all on the very same day. A written record spanning three weeks, complete with specific dates and named parties, has been gradually made public, casting the narrative of "AI conquering the holy grail of mathematics" in a far more complex and murky light.
Why the Navier-Stokes Equations Matter
The Navier-Stokes equations describe how fluids move — water, air, weather systems, blood in vessels, even the flow of a bottle of laundry detergent. Engineers use these equations every day to design aircraft wings and run weather simulations.
But the question of "existence and smoothness" has remained unresolved: can these equations, under their own internal logic, drive a perfectly smooth, well-behaved fluid to a point in space where its velocity becomes infinite — a so-called "blowup" or "singularity" — in finite time? For ninety years, no one has been able to prove or disprove this mathematically. Solving it would earn $1 million and eternal mathematical glory. Of the seven Millennium Prize Problems, only one has been solved — Russian mathematician Grigori Perelman proved the Poincaré Conjecture in 2003, yet declined both the prize money and the Fields Medal.
From Subproblem to Full Proof: A Three-Week Timeline
Understanding this controversy requires looking back at the sequence of events that preceded the announcement. Based on publicly available written records, the full picture can be clearly reconstructed.
The Prior Work of Human Mathematicians
On August 15, 2026, NYU mathematician Tristan Buckmaster and Levin Arpagu, a mathematician working at Anthropic, jointly completed proofs for two related fluid equations — the Boussinesq system and the 3D incompressible Euler equations — showing that with the addition of a smooth external force, they can undergo finite-time blowup. This can be viewed as a subproblem of the full Navier-Stokes challenge. Their work extended the methods established by Córdoba and Martínez-Zoroa.
On August 22, these proofs were formally verified in Lean — a system similar to a programming language that checks the logical steps of a mathematical argument in a mechanical, deterministic way, rather than relying on human intuition, providing a reliable endorsement of the proof's validity.
OpenAI's Involvement
On September 3, Buckmaster proactively reached out to an OpenAI mathematician, hoping to clarify the situation before internet rumors — speculating that some lab, presumably Anthropic, had solved a major problem — could spread further.
On September 6, Sébastien Bubeck, who leads OpenAI's mathematics team, spoke with Buckmaster in two phone calls. According to Buckmaster, Bubeck told him that an OpenAI model had autonomously generated an approximately hundred-page complete proof of finite-time blowup for the forced Navier-Stokes equations, and proposed giving Buckmaster authorship credit.
On September 7, Fields Medal laureate and UCLA professor Terence Tao published a blog post calling Arpagu and Buckmaster's work "a truly remarkable advance." That same day, Bubeck again requested a call; Buckmaster did not respond.
Authorship Dispute and Technical Rebuttal Erupt Simultaneously
By September 8, events had fully escalated along two parallel lines of controversy.
The Authorship Dispute: Anthropic Researcher Excluded
Buckmaster posted a four-page statement on his NYU personal webpage, accompanied by three completed papers providing finite-time blowup proofs for the incompressible porous media equation, the Boussinesq system, and the 3D Euler equations under smooth external forcing. In the statement, he alleged that when OpenAI arranged authorship credits, it deliberately excluded Arpagu because she worked for competitor Anthropic.
The head of OpenAI's mathematics team publicly dismissed the accusation as "inaccurate." Both sides stand by their accounts, and the truth remains unresolved.
The Technical Front: A Princeton Mathematician Identifies a Critical Weakness
On the same day, OpenAI formally announced its result: that the full Navier-Stokes equations can exhibit finite-time blowup. The proof was reportedly produced by an internally unreleased model running approximately 10,000 parallel AI agents over 88 hours, resulting in a 166-page manuscript that was also accompanied by a Lean formal verification.
However, Princeton mathematician Stan Palasek almost immediately identified a specific weakness in OpenAI's argument regarding the "removal of the external force" step. In brief, his rebuttal holds that the energy accumulation mechanism the proof relies upon — instabilities accumulating in separated flat bands — would dissipate energy into the fluid's viscosity faster than it could grow. In plain language: the fluid's internal resistance would cause energy to decay before it could ever "reach a singularity." Terence Tao welcomed the observation and suggested studying a simpler model to clarify where the disagreement lies.
The Capabilities and Blind Spots of Lean Formal Verification
Throughout this episode, Lean formal verification has been repeatedly invoked, and it does matter: AI can produce extremely confident "hallucinations," and a third-party system that mechanically checks a proof's validity can provide a much stronger basis for confidence in a conclusion.
But a critical limitation must be emphasized: formal verification only checks whether the mathematical logic is internally consistent within a given framework — it does not check the framework itself, nor whether the underlying premises have been correctly set up. Palasek's rebuttal targets precisely the level of premises, not the logical steps themselves — and this is exactly the blind spot that Lean verification cannot cover.
In other words, passing Lean verification does not mean the proof is correct. Mathematical history is full of proofs that initially appeared complete but were later found to contain flaws.
Does This Count as "AI Solving a Millennium Prize Problem"?
You may have missed this detail: OpenAI stated that it is not pursuing the $1 million prize, instead presenting the result as a demonstration of its model's current capabilities. This is entirely understandable — for a company already valued at over a trillion dollars and preparing for an IPO, a million dollars is trivial; the signal that "our models can do frontier mathematics" is worth far more than the prize money.
Furthermore, under the Clay Institute's rules, even if Palasek's rebuttal were fully resolved, a solution would still need to be published in a recognized mathematical journal, sustained for two years, and achieve broad acceptance within the mathematical community before any prize could be awarded. This lengthy process means that even if OpenAI's proof holds up, any official recognition of "AI solving a Millennium Prize Problem" remains far off.
The Facts Currently on Record
Setting aside speculation, what has actually been documented as of September 8 comes down to the following:
- Three papers verified by Buckmaster and Arpagu, dated between August 15 and 22;
- A series of phone call records with the head of OpenAI's mathematics team, on September 3, 6, and 7;
- A public authorship dispute;
- An internally unreleased proof announced on the same day as the authorship dispute;
- A specific technical rebuttal that has been publicly named and which no one has yet resolved.
Events are still developing rapidly. We await the final outcome of the Lean verification, whether Palasek's rebuttal holds, and the ultimate resolution of the authorship question. Whatever the outcome, this controversy has clearly revealed a challenge that is fast approaching: as AI begins to touch the outermost frontiers of human knowledge, academic attribution, result verification, and trust mechanisms will all face unprecedented tests.
Related articles
LangChain + MCP: From Core Concepts to Agent Tool Calling in Practice
Learn how LangChain and MCP work together — covering LLM tool calling, Agent architecture, and conversation history management to build real-world AI applications.
Probabilistic Machine Learning: Why It's the Cornerstone to Unlocking the ML Black Box
Without probability theory, ML is always a black box. This article explores why probabilistic foundations are essential for understanding machine learning algorithms, Bayes' theorem, MLE, and more.
Optimization Pitfalls in Self-Evolving LLM Agents: Value Concentration and Budget-Splitting Problems
HARNESSEVO research reveals 3 key LLM agent harness optimization findings: value concentrates in reflection/control slots, uniform budget splitting is harmful, and credit assignment must precede structured evolution.