FULL STORY

Caltech's AI Mathathon: From Launch to Backlash and Sponsor Exit

Caltech undergrads announced a research-level AI math hackathon backed by OpenAI and Anthropic. After hundreds of mathematicians signed a protest letter, OpenAI withdrew sponsorship and the organizers apologized and restructured the event.

2026-09-05 ~ 2026-09-13 · 7 episodes · 58 posts

Episode 1 · Caltech Hosts World's First AI Math Hackathon with Anthropic and OpenAI (2026-09-05, 23 posts)

Caltech announced the Mathathon for Oct 30–Nov 1, 2026, billed as the world's first research-level AI math hackathon, co-hosted by Anthropic and OpenAI and sponsored by roughly 30 organizations including DARPA expMath, Cognition, Cloudflare, Modal, Mercor, Prime Intellect, a16z and Y Combinator. Around 100 selected teams will solve, understand and present an open math problem over 40 hours, with over $2 million in compute credits and access to frontier models, ending with oral defenses before a panel of top mathematicians. Applications are now open. This is the first large-scale joint stress test of AI-assisted research-level mathematics by leading AI labs and a university.

Confirmed

  • The event runs three days: Friday workshop on AI agents hosted by Anthropic; hackathon-style problem solving Saturday and Sunday; oral presentations and defense before expert judges Sunday evening.
  • Participants include IMO/IOI medalists (organizer sathvikr says the pool includes 10 medalists) and Annals-level mathematicians; about 100 teams are recruited globally.
  • Prizes come in two waves: initial awards from judges for the most promising teams, with a second wave in early 2027.
  • The main track—solving and explaining an open math problem—is judged by independent academics with no sponsor involvement; sponsors run separate tracks with pre-set problems ranging from number-field sieve algorithm improvements to superalignment.
  • The organizers stress three principles: fairness (admission by math ability, travel covered), transparency (all AI conversation logs recorded and open-sourced), and understanding (teams must orally present and answer judges' questions).
  • Per sathvikr, organizers also plan a lasting responsible-use policy beyond the hackathon: convening pro-AI and anti-AI mathematicians to talk openly and co-design the rules.

Why it matters

  • Leading AI labs are joining a university for the first time to throw frontier models at open math problems at scale—a concentrated stress test of AI-assisted research-level mathematics.
  • Full recording and open-sourcing of AI conversations provides auditable public data on AI's real role in mathematical discovery, while independent judging of the main track reduces conflict-of-interest concerns.
  • Involving mathematicians on both sides of the AI debate in rule design addresses the math community's unease and offers a replicable governance template for responsible AI use in mathematics.

3 more related posts →

Episode 2 · Caltech to Host World's First Research-Grade Math Hackathon (2026-09-07, 4 posts)

A team led by Sathvik Redrouthu, together with Anthropic and OpenAI, will host the world's first research-grade math hackathon at Caltech from October 30 to November 1, 2026, bringing 100 teams to tackle an open math problem in 40 hours with $2 million in compute. All AI conversations will be open-sourced for academic review to ensure fairness.

Episode 3 · Hundreds of Mathematicians Sign Open Letter Against Caltech's AI Mathathon (2026-09-10, 9 posts)

On Sept 10, math researcher Quasilocal launched an open letter opposing Caltech's Mathathon hackathon scheduled for Oct 30. The event is organized by undergraduates: 100 teams will have 40 hours to attack open math problems using frontier LLM prompts, producing AI-generated "Lean certificates," with over $2 million in compute credits from Anthropic and OpenAI. Signatures grew from nearly 500 on Sept 10 to 700+ on Sept 11, including four Fields medalists. Signatories argue the event is effectively advertising disguised as research and that the $2M sponsorship amounts to research misconduct; critics inside and outside mathematics have pushed back forcefully.

Confirmed

  • The letter circulated as a Google doc, signed by PhD students, postdocs and faculty from Caltech, Princeton, MIT, Harvard, Berkeley, Cambridge and others, plus senior researchers at the Rényi Institute and four Fields medalists.
  • The Mathathon is undergrad-organized, with 100 teams, a 40-hour limit, LLM prompting on open problems, Lean formalized certificates, and over $2M in compute credits from Anthropic and OpenAI.
  • Core allegations: companies using massive compute to scoop open problems and publish first, "destroying valuable mathematical research directions for advertising," and flooding the research ecosystem with junk results; the $2M sponsorship is labeled research misconduct.
  • The letter also warns students that participation could damage their future academic reputations.
  • Mathematician Elliot Glazer publicly slammed the letter, saying 1500 scholars (his figure, inconsistent with other counts in the posts) decided to ruin a new three-day event for undergraduates, and that lines like "it's just an event" or "they can learn these skills elsewhere" could excuse any spiteful behavior.

Unconfirmed

  • The initiator's handle appears as both Quasilocal and Quasilierra across posts; exact spelling unverified.
  • Glazer's "1500 scholars" figure and the referenced "Navier-Stokes incident" appear only in his comments, with no primary-source corroboration.

Why it matters

  • The episode exposes strong resistance within mathematics to AI companies entering pure-math research: LLM-scale "scooping" of open problems could reshape existing norms of priority and credit.
  • Signature growth from 500 to 700+ with Fields medalist backing shows real resonance among top mathematicians.
  • The debate extends beyond mathematics and is far from one-sided: economist Ben Golub mocked the letter's logic—summarizing it as "helping young people access this technology is callous, because we might stigmatize them"; commenter Rabadash2 rejected the inference that "research math no one can read is worthless" and mocked non-mathematicians weighing in; Glazer accused boycotters of suppressing a student event. The divide centers on whether AI in math research is opportunity or stigma.

Episode 4 · OpenAI Pulls Caltech Mathathon Sponsorship After Mathematicians' Open Letter (2026-09-11, 13 posts)

On September 11, OpenAI officially announced it would withdraw its sponsorship of Caltech's Mathathon competition after considering concerns raised by members of the mathematics community, and notified the organizers. The decision followed an open letter by mathematicians criticizing AI-generated math content as "slop," signed by several prominent scholars including Fields Medalists.

Confirmed

  • Per Business Insider (relayed by @exophades), mathematicians published an open letter criticizing AI-generated mathematical content as slop, citing concerns about AI eroding the mathematical research ecosystem.
  • Per a Reddit post (@YakFull8300), the letter was signed by leading figures including 2018 Fields Medalists Peter Scholze and Alessio Figalli, and 2022 winner Hugo (post truncated; see original letter for the full list).
  • Per @ns123abc, the Mathathon had participants use large models on research-level math problems and was sponsored by multiple AI companies; the letter was signed by current and former Caltech mathematicians.
  • OpenAI (via @danintheory) confirmed the withdrawal, acknowledging that rapid AI progress in math is disruptive and expressing a desire for further dialogue with the community.
  • Background: per @soumitrashukla9, Caltech undergraduates had planned to use OpenAI-provided AI credits to attack open math problems — a generational clash that sparked the protest.
  • Comments under @TinfoilTricorn noted the event lacked remote participation options, excluding those unable to attend in person.

Why it matters

  • This is a landmark case of an AI company withdrawing sponsorship after collective community pushback, with Fields Medalists among the protesters — revealing a generational split in mathematics over AI's role: younger researchers want AI-assisted breakthroughs while senior mathematicians fear low-quality AI content polluting the field.
  • OpenAI's choice to withdraw and engage in dialogue offers a template for how AI companies and academia may interact going forward.

Episode 5 · Caltech's AI Math Competition Mathathon Revamps Rules After Backlash (2026-09-11, 4 posts)

Caltech undergraduates' Mathathon, which planned to give 100 teams compute credits to tackle open math problems with frontier AI models in 40 hours, drew an open letter criticizing incentives for rushed results. Organizers responded by promising responsible AI math experiments, changing the format and adding a six-month research period.

Episode 6 · Mathathon Organizers Redesign Event After Math Community Backlash (2026-09-11, 2 posts)

After an open letter protesting the Mathathon AI hackathon, organizers apologized and are redesigning the event with the math community, dropping the hackathon format and adding a six-month research period.

Episode 7 · Math Community Split Over AI's Role in Mathematical Research (2026-09-11, 3 posts)

A Caltech Mathathon sparked debate in the math community, with supporters saying AI-free math research is over and critics questioning whether senior mathematicians' resistance to AI-assisted problem solving is justified.