|
Getting your Trinity Audio player ready...
|
Two weeks ago, OpenAI announced that an internal AI system had produced a formal proof addressing the Navier-Stokes existence and smoothness problem, one of seven Clay Mathematics Institute Millennium Prize Problems that each carry a $1 million award for a correct solution. The announcement immediately set off debate across the mathematics community, not just over whether the proof was correct, but over whether it actually solved the problem the Clay Institute originally posed at all.
According to Scientific American, OpenAI’s proof was generated using an internal large language model working through a 166-page, Lean-verified argument that establishes a finite-time blowup, meaning a point where the equations would allow fluid flow to become infinitely fast, something physically impossible in the real world. The Navier-Stokes equations describe how fluids move, and mathematicians have spent decades trying to determine whether the equations can be trusted to behave sensibly under every condition or whether situations exist where they break down entirely. OpenAI’s system reportedly used roughly 10,000 parallel agents working over 88 hours, consuming approximately 130 billion output tokens in the process, according to figures reported by DataCamp.
The catch, and the source of the current controversy, involves a specific piece of the Navier-Stokes equations that is technically optional: an external force, such as gravity, that can either be included or left out depending on how the problem is framed. The version the Clay Institute actually poses for its $1 million prize does not include this external forcing term, but OpenAI’s proof relies specifically on a forced variant of the equations, one that many mathematicians consider disconnected from the physically realistic scenario the original problem is meant to capture. In effect, according to Scientific American’s reporting, the AI system identified and exploited a loophole in how the question could be framed rather than solving the version of the problem the prize committee actually intended.
That distinction matters considerably to the mathematics community, and it appears to matter to OpenAI as well. The company has explicitly stated it does not intend to claim the $1 million prize for this result, instead framing the achievement as evidence of how rapidly its models’ mathematical reasoning capabilities are advancing rather than as a genuine claim on the Millennium Prize itself. As of September 2026, the Clay Mathematics Institute still lists the original, unforced version of the Navier-Stokes problem as an active, unsolved challenge, according to background compiled on the problem’s official documentation.
Complicating matters further, three mathematicians posted their own proof just last Thursday demonstrating that OpenAI’s forced-variant approach can never be extended to solve the actual, unforced problem the Clay Institute poses. That finding effectively closes off any path from OpenAI’s current result toward the version of Navier-Stokes that would actually qualify for the prize, barring some entirely new mathematical idea nobody has yet identified. In other words, according to Scientific American, the loophole OpenAI’s system found leads to a genuine dead end rather than a stepping stone toward the harder, real problem.
The story took a more contentious turn when NYU mathematician Tristan Buckmaster raised separate concerns about how OpenAI’s system arrived at its results. Buckmaster, who had spent roughly a year working with collaborator Levent Alpöge on a related approach to the Euler equations, a simpler, frictionless cousin of Navier-Stokes, using AI models from OpenAI’s rival Anthropic, alleged that OpenAI’s system may have drawn on his and Alpöge’s private chat logs and uploaded drafts. Sebastien Bubeck, a senior OpenAI researcher, rejected that characterization directly, telling Scientific American the company did not use their prompts, models or proof, and separately called Buckmaster’s broader account false and inflammatory. Buckmaster further alleged that when he indicated he intended to make these concerns public, he was discouraged from doing so, though OpenAI has denied accessing any private or unpublished work belonging to Buckmaster and Alpöge, and the underlying private conversation between the parties remains genuinely disputed.
Diego Córdoba, one of the mathematicians whose earlier theoretical work laid the groundwork for the blowup technique in question, offered a notably cautious reaction when asked whether the prize had effectively been claimed, saying that such an outcome would come as a significant surprise to him. That skepticism from a mathematician closely connected to the underlying technique underscores just how unsettled the mathematics community’s assessment of OpenAI’s result remains, even weeks after the initial announcement generated widespread attention.
The episode arrived amid a broader wave of AI labs racing to demonstrate mathematical reasoning capabilities through high-profile problem-solving claims. According to DataCamp’s reporting, OpenAI’s Navier-Stokes effort was reportedly launched in part after CEO Sam Altman heard that rival Anthropic’s models had made progress on a major math problem, prompting an internal push to test whether OpenAI’s own systems could achieve something comparable. That competitive dynamic, paired with the significant compute expense involved, estimated by some outlets at tens of millions of dollars, far exceeding the $1 million prize itself, has led some observers to argue that corporate prestige and demonstrated model capability, rather than an actual bid for prize money, was the primary motivation behind the effort from the outset.
Whether OpenAI’s proof stands as a genuine, if narrower, mathematical achievement worth taking seriously, or primarily illustrates how AI systems can technically satisfy a problem’s literal wording while missing its intended substance, remains a live and unresolved question within the mathematics community. What is clear is that the original, unforced Navier-Stokes problem, the one the Clay Institute actually offers $1 million to solve, remains just as unsolved today as it was before OpenAI’s announcement, and the newly published proof showing the forced approach cannot be extended further suggests that closing the actual gap will require a fundamentally different mathematical strategy. Continuing coverage of how AI systems are being applied to unsolved problems in mathematics and science is available on Business Tech. Additional detail on the controversy is available through Scientific American’s original reporting, and background on the Millennium Prize Problems themselves can be found through the Clay Mathematics Institute’s official site.
