On September 4th, we announced Mathathon, an event allocating compute credits across 100 teams of mathematicians tasked with solving and presenting an open problem in 40 hours. As of September 10th, we have received over a thousand applications from undergraduates, Ph.D. students, postdocs, and faculty at top research institutions in both the U.S. and abroad.
A recent open letter raised serious concerns about whether the event could incentivize rushed, poorly understood mathematics, place verification burdens on the broader community, and amplify unhealthy incentives around AI-generated results. We think several of these concerns are valid, and our original public description did not adequately explain or, in some cases, adequately address them. We have therefore updated our public description and added additional requirements to second-round submissions.
We are writing this statement to clarify what we have done to address the community’s concerns. If successful, Mathathon will not be a list of 100 problems with convoluted solutions and occasional Lean verifications. Instead, it will be an experiment in what responsible AI usage for mathematics could look like, with the goal of preserving the culture of mathematics which we and so many others call home.
Who are we, and what do we believe?
The primary organizers of Mathathon are undergraduates, the majority of whom plan to pursue theoretical mathematics for the foreseeable future. We are personally invested in what the human mathematical experience should look like in an AI future. As such, we are committed to conducting our event with respect for existing academic communities. At the same time, we believe that preserving the spirit of mathematics requires experimentation with resource allocation and incentive structures. Mathathon is one such experiment.
Throughout the organizational process, we have consistently sought guidance from members of the mathematical community. They include graduate students, junior faculty members, and tenured professors. They include both AI-positive and AI-skeptic mathematicians. These discussions informed our responsible AI policies, including the 6-month verification period and the educational tracks. After the open letter and additional discussions with our faculty advisors, we are now making arXiv publication an explicit requirement for the verification period.
Mathathon begins with a 40-hour period from October 30th to November 1st where teams use provided credits to produce an initial result. We will award additional AI credits to the most promising and well-explained solutions.
However, we understand that it is unrealistic to expect participants to fully understand and internalize a novel result in just 40 hours. This is why Mathathon is structured in two rounds. Participants who produce results sufficient to warrant consideration by the math community—as determined by our judges—will progress to the second round. The second round, also known as the verification period, will take place over the next 6 months during which participants will produce an arXiv pre-print, a video presentation, and an auto-formalized solution. Our judges will evaluate them throughout this period, and we are awarding more prizes at the end. We are also hosting the educational track, in which participants produce course notes, video explainers, or more elegant proofs of existing theorems.
This layout is inspired by existing structures in math academia and educational prizes, such as REUs, workshops, the Breakthrough Junior Challenge, and the Summer of Mathematical Exposition.
Against the “Slopification” of Mathematics
Firstly, the majority of our applicants so far are PhD students who have a proven record of quality mathematics publications. Secondly, guiding our decision to allow applications from all levels of the mathematical community—including undergraduates—is our belief that, to progress as a field, we must trust each individual to use resources responsibly. In particular, we are placing trust in our participants to refrain from releasing/uploading results until they have fully internalized and verified their work. We believe that the incentive structures of the competition support this goal. Participants are rewarded not for being the first to solve a problem, but for explaining their results, motivations, and techniques.
We cannot guarantee that all work produced at this event will meet the mathematical community’s standards, but we are implementing the measures we can to mitigate that risk, and we welcome any community suggestions to further reduce that possibility.
Intrusion of Corporate Interest
We are grateful to have received over 2M compute credits from OpenAI and Anthropic along with support from DARPA expMath, Cognition, and other industry partners. Compute is a crucial catalyst for problem discovery, resolution, verification, presentation, and ultimately, the success of Mathathon. That being said, we understand that significant portions of the scientific community are concerned that the goals of major AI corporations do not necessarily align with the aims of scientific research. We see Mathathon as a step towards critically engaging these parties. Besides the goals of the event itself, we plan to host debates, fireside discussions, and panels between AI critics, supporters, and those in between.
Care for Young Mathematicians
As young mathematicians ourselves, we are well aware of the stress that the uncertain future of mathematics and academia bring. If anything, Mathathon aims to help early-career mathematicians in a time where access to compute resources asymmetrically favors established, well-funded researchers. Our 6-month second round encourages motivated participants to take their problem seriously, study it deeply, and leave with a better understanding of the material, just like existing REU programs or directed reading programs. With a steepening decline in Ph.D. applications for theoretical fields of study, events like Mathathon encourage young people to stay excited about mathematics, engage their interests with modern tools, and maintain hope in a confusing time. As for concerns that association with AI companies will negatively impact the reputations of theory-minded participants, we respect our applicants and the mathematical community’s ability to make the right decisions.
Misinformation about Research Mathematics
We take full responsibility for any public statements we put out that were wrong or misleading. We have retracted the question “what is the role of a mathematician when AI can solve conjectures faster” from our website. It was intended as a statement about a future possibility, but we have since realized that it could be interpreted to mean the present. To prevent this from recurring, we will ask our faculty advisors to review any future statements. It’s our first time running an event at this scale, and we are grateful for all the feedback we have received.
Receiving Feedback from the Community
To the best of our knowledge, we received three proposals from signatories of the letter before it was published. We responded as follow:
- Add a track where participants create educational content about an existing theorem.
- We are hosting this track and allocating to it a similar amount of prizes as our main track.
- Require participants to publish their results on arXiv.
- Participants should only publish after they have verified their results, checked their citations, and labelled their AI usage. We encourage them to take a few days after Mathathon to do this. We would require it if we can, but we can’t control their actions after the event. [Note: This policy has been updated to require publication in the verification round.]
- Invite a specific signatory to give a 10-to-30-minute speech during the opening ceremony.
- We aim to finalize our speaking arrangements by the end of September. We haven’t assigned any speech / workshop slot to anyone. We will host panel discussions between AI-positive and AI-critical mathematicians. We welcome her as a speaker. But we cannot determine at this time whether she’ll be a headline speaker.
Mathathon is an experiment. We hope to learn whether substantial AI resources can be incorporated into mathematical research while preserving the values that make mathematics worth doing: understanding, attribution, openness, mentorship, and verification. We greatly appreciate the time taken by anyone who has given feedback regarding Mathathon, and we will keep working with them to improve this event. We will reach out to signatories of the letter to have a conversation about their concerns, and we will respond to feedback received through our email.
All organizers have signed the Leiden Declaration.
Received 10 September 2026.
Add to the discussion