Caltech students organized a ‘Mathathon’ sponsored by Anthropic and OpenAI. Then came the backlash

“40 hours, $2M+ AI credits, solve an open problem.”

So read the splashy announcement for the “Mathathon” at the California Institute of Technology in Pasadena. Dozens of teams would work to solve vexing math problems with the help of artificial intelligence as part of “the first hackathon ever devoted to research level mathematics,” according to the early September post. AI juggernauts Anthropic and OpenAI were set to sponsor the two-day competition.  

The backlash came swiftly. Critics objected not only to using AI to tackle open problems, but also to the involvement of AI companies whose models and research practices they have criticized. 

Within days, current and former Caltech mathematicians published an open letter calling on the undergraduate organizers to suspend the event. The Mathathon, they argued, would only fuel the creation of “slop mathematics” and serve as public relations for the companies.

Instead of canceling the event, Caltech student organizers responded as any mathematician would — they tried to solve the problem. 

The undergraduates redesigned the competition. Instead of trying to solve open mathematical problems, teams will now aim to develop alternative proofs for existing, yet unsatisfying solutions. The November event will also no longer be sponsored by the companies that develop proprietary AI models.

“We came to the decision that we want to put up an event that unites the math community, rather than divides it,” Brian Zhao, one of the Mathathon organizers and a junior studying applied mathematics at Caltech, told EdSource.

The controversy over the Mathathon event laid bare the growing tensions unfolding in the mathematics community over what role AI should play in advancing the field. 

AI has already helped to solve elusive math problems. OpenAI announced this month that it used an internal AI model “significantly more capable” than the latest publicly available models to solve one of the remaining six Millennium Prize Problems. But mathematicians have warned about AI’s negative impact on mathematics students, and criticized these companies’ approach to solving high-profile problems.

“To put it bluntly, AI companies are engaging in research misconduct,” wrote the authors of the open letter, which gained the signatures of 2,038 academics as of this week.

‘That’s not the culture of mathematics’

Earlier this summer, Zhao and his roommate came up with their idea of a fun weekend: lock themselves in a basement and try to solve math problems for 40 hours straight. He was inspired by hackathon-style competitions in which programmers and designers compete to build hardware or software under deadline.

The idea of using AI to try to solve open math problems was two-fold. With mathematics undergoing one of the biggest changes in decades because of AI, the Caltech undergrads wanted to organize an event to experiment with AI and better understand how the field may evolve by the time they start graduate school. 

Also, the students hoped that advertising the event with flashy statements — like suggesting teams would have access to $2 million in AI credits — would attract high-profile professors as judges and big lab sponsors. 

“What is the role of a mathematician when AI can solve conjectures faster?” Mathathon organizers initially asked on the event’s website. 

The AI-infused premise of Mathathon drew the ire of current and former Caltech mathematicians.

“I found it pretty outrageous, taking $2 million worth of sponsorships from OpenAI and Anthropic (and) partnering with these companies that have been, in my opinion, and many other mathematicians’ opinion, very disrespectful to our field,” said Yujin Kim, a postdoctoral fellow in Princeton University’s math department. 

Kim had been critical of OpenAI’s approach to solving elusive mathematics problems in viral social media videos. The companies’ sponsorship of the event “felt like aligning with the empire,” Kim said.

“I can’t imagine (the Mathathon) would go any way other than just essentially button pressing and prompting,” Kim said. “That’s not the culture of mathematics. That’s not how we operate.” 

After the backlash, OpenAI pulled its support from the event. 

“We recognize that the rapid progress of AI in mathematics is disruptive,” Dan Roberts, OpenAI’s research lead, said in a social media post after the open letter was published. A spokesperson for Anthropic declined to comment.

The updated Mathathon competition will allow participants to use AI, but highly encourage them to use open-source tools that can be modified by the public instead of proprietary models run by private entities like OpenAI.

“We are proud of all of our students for the passion, drive, and maturity they have shown throughout this development,” Caltech spokesperson Emily Velasco said in a statement. “The spirit of open inquiry is central to the scientific process, and we hope to see a continued exchange of ideas as the math community charts its course forward.”

Zhao said he and his fellow organizers are determined to move forward with hosting the event. He said organizers hope the Mathathon will allow students to discover ways to use the technology responsibly, and build a bridge between AI developers and the math community. 

AI can act like a shortcut to solving math problems

AI has conjured much debate in the math community, some of which has been overdue, said Dimitri Shlyakhtenko, a UCLA math professor who has followed Mathathon developments but did not sign the open letter.

From the outside, it may look like AI is upending the field by solving high-profile problems that have stumped mathematicians for decades. But the reality is more nuanced, Shlyakhtenko said.

“It is very rare that some result in mathematics is obtained out of nothing,” he said. “It usually builds on a mountain of other works that happen, and usually there’s a community of people that care about a particular set of problems.”

Mathematics problems aren’t about the destination of finding a solution, he said. Rather, mathematicians develop new techniques and understandings along the journey to solve a problem. He likened it to hiking to a waterfall. The end goal is to see the waterfall, but the things you discover along the journey are also quite interesting and of value. 

With AI, he said that “what we’re seeing right now is that it is sometimes possible to take a kind of a direct shortcut.” 

Some mathematicians worry that using AI to solve problems robs training opportunities from younger mathematicians. 

Training problems provide opportunities for graduate students to learn skills and approaches to mathematics, but “if those just get solved, then the motivation to do it is kind of gone, right?” Shlyakhtenko said.

Over the past several months, as the Mathathon drama unfolded, Zhao said he has learned two important lessons: Be ambitious and be humble.

Maintaining the lofty goal of the event — of trying to help unravel the thorny debate of AI’s role in mathematics — has helped the team persist in the face of challenges, Zhao said. He added that it also developed the event into something more than just a fun, competitive weekend. The conversations with mathematicians around the world about the event have also taught him to be wary of flashy statements. 

“If we just present ourselves honestly as undergrads who are curious about AI, we actually attract more sympathy,” Zhao said. “We actually attract more mathematicians who are willing to help us.”

Related Reading

College-bound Bay Area high school students say AI is useful, and a source of anxiety

September 22, 2026

Teens use AI for academic support even though they see the dangers, survey finds

August 18, 2026

California State University renews controversial systemwide contract with OpenAI

May 21, 2026

This story was originally published by EdSource.