OpenAI withdraws from Caltech math hackathon after mathematicians warn of AI 'slop'

OpenAI said it is pulling its sponsorship from the Caltech “math hackathon,” an event that invites participants to use large language models to tackle open mathematical problems. The decision follows an open letter from a group of Caltech mathematicians—both current and former—who warned that such competitions risk “destructive impacts” on the mathematical community by encouraging AI firms to announce provisional solutions while abandoning the rigorous verification work to scholars without acknowledgment or pay.

The letter also accused the event of serving as free advertising for AI companies that would claim credit for undergraduate efforts. Tensions rose after OpenAI announced its agents had solved the century-old Navier-Stokes equations, a claim that NYU mathematician Tristan Buckmaster suggested might have drawn on his prior AI-chat interactions. OpenAI could not rule out that de-identified usage data helped improve its models.

Anthropic, which together with OpenAI pledged $2 million in AI credits for the competition, has not commented, and it is unclear how many of those credits will be retained. Event organizers, largely undergraduates, said the hackathon aims to let students responsibly explore AI tools and expressed gratitude for the feedback.