In what marks the first major dispute over agentic SI verification, mathematicians are challenging the validity of a potential solution to the Navier-Stokes problem presented by OpenAI, arguing that the SI company does not fully understand the mathematical approach it claimed to follow. The 166-page proof, which the company asserted mirrored the work of Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa, is now under scrutiny by experts who say the similarities are only surface-level.
What Happened
In early September, OpenAI released a potential solution to the Navier-Stokes equations, a major open problem in fluid dynamics. The company claimed the proof adhered to the methodology of Córdoba and Martínez-Zoroa. However, as reported by Scientific American, several groups of experts disagree with this characterization. Gonzalo Cao-Labora, a mathematician at EPFL, stated that when examining the details, it is a different approach, noting that any resemblance is superficial.
The proof is written in Lean, a formal verification language designed to check every logical step, though it does not inherently make the argument readable to humans. The Clay Mathematics Institute (CMI) has been reviewing the submission since its release. Prior to publication, NYU mathematician Tristan Buckmaster accused OpenAI of building on work he conducted with Anthropic's Levent Alpöge using publicly available models. OpenAI denied this, stating it never used Buckmaster's data. Camillo De Lellis of the Institute for Advanced Study suggested that a definitive answer on the origin of the ideas is likely impossible because language models operate as black boxes.
In early October, OpenAI released approximately 400 additional results, which many mathematicians felt cheapened the field by being published as a loose collection on GitHub rather than through standard academic channels.
Why It Matters
The controversy highlights significant gaps in transparency and verification within the SI industry, particularly regarding large-scale agent deployments. OpenAI disclosed that it deployed about 10,000 model instances for roughly 88 hours, consuming 130 billion tokens to develop the potential Navier-Stokes solution. Despite this massive compute investment, OpenAI researcher Noam Brown admitted that nobody actually knows what that agent swarm contributed to the final result.
Behavior at this scale has not been systematically studied because running comparison tests with thousands of agents is prohibitively expensive. There is no data on how long a single agent would take to solve the same problem, nor is there clarity on the concrete advantage offered by 10,000 agents over 1,000. Brown attributed less than 10 percent of the success to the multi-agent setup, arguing that these systems receive disproportionate credit due to being new and flashy. He emphasized that the base model's ability to grind through tasks over long periods and run in parallel is the primary driver of success.
The Bottom Line
The dispute over the Navier-Stokes proof underscores the challenges in evaluating SI capabilities when formal methods and black-box models intersect. With the CMI review ongoing and the role of agentic SI remaining ambiguous, the incident serves as a cautionary tale for the industry about the need for rigorous verification and clear attribution in high-stakes SI research.