Mathematicians Challenge OpenAI's Navier-Stokes Proof Methodology
Experts dispute OpenAI's claims about its proof source and the value of its multi-agent computing strategy.
Key points
- Mathematicians state the proof uses a different approach than OpenAI described.
- OpenAI admits it is unclear what value 10,000 agents added to the solution.
- The Clay Mathematics Institute is currently reviewing the 166-page Lean code.
OpenAI submitted a 166-page potential solution to the Navier-Stokes problem in early September, but mathematicians now argue the company does not fully understand its own work. Experts contend that the methodology differs significantly from what was publicly claimed, casting doubt on the origin and validity of the approach.
According to Scientific American, several groups of specialists dispute the connection to the work of Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa. Gonzalo Cao-Labora of EPFL stated that the similarities are only surface-level and that the actual details represent a different approach entirely. The proof is written in Lean, a formal verification language that checks logical steps but remains difficult for humans to read directly.
Review and Attribution Disputes
The Clay Mathematics Institute has been reviewing the submission since its release in September. Before publication, NYU mathematician Tristan Buckmaster accused OpenAI of building on work he and Anthropic researcher Levent Alpöge had done with publicly available models. OpenAI denied using Buckmaster's data, but Camillo De Lellis of the Institute for Advanced Study noted that a definitive answer is likely impossible because language models are black boxes.
In early October, OpenAI released roughly 400 additional results on GitHub. This loose collection angered many mathematicians, who felt the publication style cheapened their field. The math community struggled to keep up with the volume of output, which further complicated the peer review process and added confusion to the ongoing debate regarding the proof's origins.
The Multi-Agent Uncertainty
OpenAI deployed approximately 10,000 model instances for roughly 88 hours and 130 billion tokens to develop the solution. However, researcher Noam Brown admitted that nobody knows what that massive agent swarm actually contributed to the result. Behavior at that scale has not been studied because running systematic comparison tests with thousands of agents is too expensive.
Brown attributed less than 10 percent of the success to the multi-agent setup. He suggested these systems receive disproportionate credit because they are new and flashy. He argued that the base model itself is what matters, as it can grind through tasks over long periods and run in parallel. No one has tested how long a single agent would take to solve the same problem, making it nearly impossible to determine the concrete advantage of using 10,000 agents over 1,000.
What should you do?
Security and technology professionals should monitor the Clay Mathematics Institute for official validation results. Organizations relying on AI for complex technical tasks should verify outputs independently rather than trusting black-box explanations. Teams should scrutinize claims about multi-agent efficiency, as the incremental value of large agent swarms remains unproven.
What to do and how to stay safe: OpenAI
- Verify AI-generated technical proofs using independent formal verification tools or human experts.
- Assess whether multi-agent setups provide measurable advantages over single-model processing for your use cases.
- Maintain strict attribution records when using publicly available models for research to avoid plagiarism disputes.
- Monitor official channels like the Clay Mathematics Institute for definitive rulings on high-profile AI submissions.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Did OpenAI admit to using Tristan Buckmaster's work?
OpenAI denied using Buckmaster's data, though experts say a definitive answer is likely impossible due to the black-box nature of language models.
What language was the Navier-Stokes proof written in?
The potential proof was written in Lean, a formal verification language that checks every logical step but is not easily readable to humans.
How much of the success does Noam Brown attribute to the multi-agent setup?
Noam Brown attributes less than 10 percent of the success to the multi-agent setup, crediting the base model instead.
