AI firms clash over claimed solution to century-old Navier-Stokes puzzle

George Gabriel Stokes, one outlet-century Irish mathematician who helped formulate the Navier-Stokes equations, gave his name to one of the seven Millennium Prize Problems. OpenAI recently released a 166-page document claiming its large-language-model agents produced a "blow-up"—a situation where fluid velocity would become infinite—demonstrating that the equations do not always hold. About 10,000 agents worked for 88 hours to construct this counter-example, prompting mathematicians to examine the proof and its accompanying code.

Anthropic, another AI firm, was reported to have collaborated with academics on the same problem, and allegations have surfaced that it passed insights to OpenAI, which it rejects. DeepMind’s co-founder Demis Hassabis, speaking at an AI summit in Dublin, argued that his company prefers to work alongside the mathematics community rather than claim solitary answers. The episode has ignited a broader debate about how AI can contribute to fundamental research and how such claims should be validated.