OpenAI pulls Caltech funding as feud with top mathematicians boils over

What happened?

OpenAI has abruptly pulled its sponsorship from a major Caltech mathematics competition, marking a sharp escalation in a growing feud between the artificial intelligence pioneer and elite academics.

The company's research lead, Dan Roberts, announced the withdrawal on Thursday following intense criticism from Caltech researchers. In an open letter, a group of current and former mathematicians at the university had accused AI firms of waging a "campaign of scientific misinformation" and producing "slop mathematics" designed to market commercial products rather than advance genuine understanding.

The sudden pullout leaves Caltech organisers scrambling to fill a funding gap for its "Mathathon," scheduled to begin on 30 October. OpenAI and its rival Anthropic had collectively promised $2 million in computing credits to fund 100 competing teams. A spokesperson for the event said OpenAI was providing half of the $20,000 in tokens allocated to each team, but insisted they are in talks with other firms and do not expect the disruption to derail the competition.

Who is involved?

The dispute is not just about sponsorship money. It represents a fundamental clash of cultures that has been building for months, culminating this week in a bitter row over a legendary mathematical puzzle.

OpenAI recently claimed it had solved the Navier-Stokes Problem, a centuries-old challenge concerning the way fluids move. The academic community quickly disputed the claim, pointing out that OpenAI's proof remains unverified.

The announcement also triggered a public fallout with New York University mathematician Tristan Buckmaster. He accused OpenAI of pressuring him to withhold credit from his collaborator, Levent Alpöge, because Alpöge works for Anthropic.

The dispute has raised deeper suspicions. Both Buckmaster and Alpöge have publicly questioned whether OpenAI used data from their research, conducted using OpenAI's Codex coding tool, to train its models and then rush out its own proof over a single weekend.

Why does it matter?

For many in the field, the episode confirms their worst fears about how Silicon Valley's "move fast and break things" ethos is colliding with the rigorous, slow-paced world of academic research.

This week, 25 Fields Medal winners—the highest honour in mathematics—signed an open letter warning that the frantic race between AI labs threatens to break the "human transmission chain" that keeps the discipline alive. When AI models spit out solutions without a proper write-up or peer review, they argue, the intellectual scaffolding of the subject is lost. Rushing these announcements, the signatories wrote, leaves "no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work of others."

There is also a growing worry about academic survival. If tech companies can use their vast computing power to monitor researchers' work and then spend millions of dollars in computing power to "scoop" them on a proof, the traditional culture of open academic collaboration will die. Instead, mathematicians warn, researchers will be forced to work in complete secrecy to protect their intellectual property.

What happens next?

OpenAI says it wants to find a way to patch things up. In his announcement, Roberts wrote that the company recognises its progress in mathematics is "disruptive" and that it wants to engage more with the community to figure out how to integrate the technology.

But mathematicians have already begun organising. The current backlash follows the Leiden Declaration in June, in which a working group of mathematicians laid out guidelines to protect their profession from being overrun by large language models. With elite academics now openly accusing tech firms of plagiarism and misconduct, the gap between the two worlds looks wider than ever.