When a group of Caltech undergraduates announced the Caltech Mathathon — billed as the first hackathon devoted to research-level mathematics — they framed it as a public good. One hundred teams, $20,000 or more in AI credits per team, 40 hours on the Caltech campus, and a single goal: make real progress on open problems in mathematics, with frontier models put in the hands of the math community rather than locked inside AI labs.
It was, on paper, a generous experiment. Sponsors including OpenAI, Anthropic, DARPA’s expMath program, a16z, Y Combinator, and Cognition signed on.
The mathematics community did not see it that way. An open letter opposing the event drew roughly 500 signatories, Caltech mathematicians announced a boycott, and critics warned the hackathon would fuel “slop mathematics” — plausible-looking but shallow AI-generated results that clutter the literature and erode the standards of proof the field runs on. On September 10, OpenAI scientist Dan Roberts announced on X that the company had withdrawn its sponsorship, citing “concerns raised by members of the mathematics community,” and had notified the organizers.
The Mathathon, at least for now, has lost its biggest backer.
The Open Letter
“AI companies see research mathematics as an advertising opportunity,” the letter said. “Over the past few months, they have advanced a campaign of scientific misinformation about the goals of mathematical research. Major AI companies are engaged in an arms race to declare themselves the first to prove prominent open conjectures. New announcements of AI-generated results appear on social media on a weekly basis. These results are often poorly communicated, and have diffuse negative impacts on the careers of human mathematicians who are concurrently proving the same results,” it added.
“Mathematicians do not only prove theorems: they also take measures to ensure that the future of a field remains healthy after the extraction of a result. They do so by sharing their results and explaining their methods to others at talks and conferences. They mentor students and postdocs so that the next generation can advance research further, and take care not to scoop other mathematicians at the last mile. In announcing new LLM-generated theorems, AI companies have not followed such research practices. The intent is clearly to extract the prestige of newly proven results rather than add value to the community in a way that safeguards its future,” the letter continued. “These companies appear guided by an unhealthy instinct to claim certain results before competitors at all costs, regardless of the collateral damage to mathematical understanding,” it alleged.
A generational fault line
The episode lays bare a fault line that has been widening all year. On one side: young researchers and AI labs who see a genuine acceleration — models that recently produced new results on ten open problems at once, resolved Erdős problems that sat untouched for decades, and posted gold-medal scores at the International Mathematical Olympiad. Even top mathematicians like Terrance Tao is using AI to experiment and try “crazier things”.)
On the other side: senior mathematicians who believe the labs are mistaking problem-solving for mathematics. Tao himself has warned that AI-powered proofs could be a net negative for the field, arguing that understanding — not the mere production of correct statements — is what the discipline is for. Peter Scholze, a Fields Medalist, has said he prefers to “ponder his mathematical ideas without use of AI.”
The Hacker News thread announcing the Mathathon captured the mood. One mathematician called it a scheme to use “professional mathematicians as (cheap?) labour for validating LLM outputs.” Another warned that teams’ failed attempts could be mined for ideas the labs would later claim as their own. A recurring question: why was Caltech’s own mathematics department not involved in organizing an event being held on its campus?
The backlash has been building for months
The Mathathon revolt didn’t come out of nowhere. It’s the sharpest flashpoint yet in a backlash that has been organizing since at least June, when the Leiden Declaration on AI and Mathematics — endorsed by the International Mathematical Union — warned that AI threatens the “characteristic values” of mathematical research: verifiable proof, proper attribution, and autonomy from commercial interests. It specifically criticized results “communicated through informal channels such as press releases,” noting that this oversimplifies media coverage and “misleadingly uses specific mathematical tasks as metrics for the general reasoning capacities of commercial products.”
That was a barely veiled reference to OpenAI, which had just weeks earlier publicized its model disproving the 80-year-old unit distance conjecture. The skepticism has only sharpened since. NYU mathematician Tristan Buckmaster publicly accused OpenAI of trying to take credit for AI-assisted work on a fluid dynamics problem that he says originated in his own use of the company’s tools. Mathematician Andreas Thom then raised concerns that nonpublic research shared with ChatGPT may have informed OpenAI’s own non-sofic groups result.
The deeper fear among senior researchers isn’t that AI will solve problems. It’s that the economics of the field will invert: proofs that no human can read, generated at trivial cost, flooding journals and hiring committees, while the slow, human work of cultivating understanding — the work careers and reputations are built on — gets devalued. Mathematics has always been a status economy as much as a search for truth. AI has, almost overnight, made the status symbols feel cheap.
OpenAI’s careful retreat
OpenAI’s withdrawal reads as an attempt to repair relationships before the damage compounds. “We recognize that the rapid progress of AI in mathematics is disruptive,” Roberts wrote. “We’re looking to engage with the math community more on the best way to integrate this technology and communicate its impacts.”
Notably, the other sponsors — Anthropic, a16z, Y Combinator — have not said they are leaving. The Mathathon’s organizers, who stressed that the event was student-led and that sponsors “don’t have a say in our decisions,” say they plan to continue.
But the episode signals something larger for the AI industry. Mathematics was supposed to be the clean, undeniable showcase for AI reasoning — a domain with objective answers and no messy real-world consequences. Instead, it has become the first field where the pushback is organized, credentialed, and effective. The labs can beat humans at solving problems. Convincing humans that’s a good thing is proving to be the harder theorem.