OpenAI's push into Millennium Prize maths problems triggers revolt by top mathematicians

What is the breakthrough?

OpenAI is closing in on proving a second million-dollar Millennium Prize mathematics problem, according to its president Greg Brockman, just weeks after the company claimed to have solved a century-old equation governing fluid mechanics.

The rapid intrusion of artificial intelligence into pure mathematics has sparked a fierce backlash from the academic community. Twenty-five winners of the Fields Medal—the discipline's highest honour—have signed an open letter expressing deep fear over a "severe misalignment" between Silicon Valley and their field.

While tech enthusiasts are celebrating what they call the dawn of artificial general intelligence, many academics worry that the flash of human genius is being systematically replaced by raw, industrial-scale computing power.

Why are mathematicians angry?

The anxiety is not just philosophical; it has devolved into accusations of academic poaching.

Tristan Buckmaster, a mathematician at New York University, has accused OpenAI of taking credit for academic research. Buckmaster and Anthropic mathematician Levent Alpöge had been using public AI tools to work on the Navier-Stokes equations—a Millennium Prize problem concerning how fluids move in three-dimensional space.

According to Buckmaster, information about their progress was passed to OpenAI. The tech giant then allegedly unleashed an unreleased, next-generation model to finish the proof, spending an estimated $22.5 million in computing power over a single week to beat the academics to the finish line.

Buckmaster also claimed that OpenAI researcher Sébastien Bubeck pressured him to remove Alpöge’s name from their collaborative work, allegedly threatening to "ruin" his career if he did not comply.

Brockman has denied these claims. He stated that the AI model OpenAI used for its proof had a training cutoff of July, well before the academics' work was made public, and insisted the company does not snoop on private developer data.

How did the AI solve the problem?

The sheer scale of OpenAI's mathematical operation is unlike anything seen in academic history.

To tackle the Navier-Stokes equations, the company did not rely on a single researcher in an office. Instead, it deployed a swarm of roughly 10,000 reasoning AI agents. Working together for 88 hours, these agents exchanged 2.7 million messages and went through 130 billion output tokens of reasoning.

The digital swarm eventually proved that smooth fluids can generate singularities in finite time, writing the entire proof in Lean, a computer language used to formally verify mathematical logic.

"Back then, Andrew Wiles spent ten years secretly working alone in the attic to prove Fermat’s Last Theorem," Brockman said, drawing a contrast with the modern era. Today, a tech giant with massive computing resources can flatten a century-old mystery in a few days.

What happens next?

The identity of the second Millennium Prize problem OpenAI is targeting remains a secret, though speculation in Silicon Valley points to the famous P vs NP problem, which governs the fundamental limits of computation and modern cryptography.

The rapid progress has divided observers. Scott Armstrong, a professor of mathematics at NYU's Courant Institute, said the mathematical "Singularity" has already begun.

But critics question whether using scarce global computing power to chase abstract mathematics prizes is the best use of resources. Many argue these supercomputers should be deployed to solve real-world crises, such as finding new cancer therapies or predicting protein mutations.

Brockman insists that advanced mathematical reasoning is the ultimate testing ground. He argued that a system capable of executing complex, error-free logical chains will eventually be migrated to molecular biology and genetics, where its life-saving potential will grow exponentially.