Mathematicians Revolt, and OpenAI Quits a Math Contest

OpenAI has withdrawn its sponsorship of the Caltech Mathathon, a competition where entrants would race to crack open research problems by prompting large language models, after mathematicians published an open letter warning the event could do lasting harm to their field. The reversal is a rare case of a frontier lab backing down in the face of pushback from the academic community it keeps saying it wants to help.

The event, due to start on October 30, had lined up Anthropic and OpenAI to hand participants some $2 million in AI credits. Then came the open letter from current and former Caltech mathematicians, who argued the contest "is likely to have destructive impacts for the mathematical community." OpenAI scientist Dan Roberts said the company had listened to those concerns and told the organisers it was pulling out.

The charge: slop mathematics

Two objections sit at the heart of the letter. The first is quality. Turning graduate students loose to generate proofs by prompting a model, the signatories argued, threatens to bury journals and preprint servers under "slop mathematics," plausible-looking arguments that no human has properly verified. As Understanding AI has reported, the discipline runs on trust that a claimed proof has actually been checked, and a flood of machine-generated candidates strains the volunteers who do that checking.

The second objection is about people. The mathematicians said an event that has early-career researchers feed problems to a corporate model, rather than collaborate on the mathematics itself, exploits them at the most fragile stage of their careers while handing the labs free publicity and training signal.

A pattern, not a one-off

The dispute lands in a season of loud claims about AI and mathematics. OpenAI said it marshalled 10,000 agents to force a proof related to a Millennium Prize problem, a claim that itself drew a credit dispute, and Anthropic reported that Claude produced a machine-verified proof of Fermat's Last Theorem in Lean. The Mathathon backlash is the same argument from the other direction: not whether the machines can do mathematics, but whether the way labs are deploying them respects the community that has to live with the results.

It is worth separating the fears from the record, as we did when asking whether AI has made mathematicians obsolete. Verified formal proofs are real progress. Contests that reward volume over verification are a different thing, and this time the mathematicians said so loudly enough that a sponsor listened.