Twenty-five winners of the Fields Medal, mathematics' highest honour, signed a declaration on 11 September 2026 stating that "the goals of the AI companies and the goals of the mathematical community are severely misaligned." It arrived one day after OpenAI withdrew its sponsorship of the Caltech Mathathon, an AI-assisted math hackathon, following a separate open letter from Caltech mathematicians that accused AI companies of "research misconduct."
Key facts
- 25 Fields Medallists signed the declaration, among them Terence Tao, Peter Scholze, Maryna Viazovska, James Maynard and Maxim Kontsevich; more than 800 researchers had added verified endorsements by the evening of 11 September.
- On 10 September, OpenAI scientist Dan Roberts announced that OpenAI was withdrawing its sponsorship of the Mathathon, after an open letter that had 771 signatories at publication.
- The declaration names no company. The Caltech letter names both Anthropic and OpenAI, which were supplying participants with $2 million in AI credits.
- Primary sources: the declaration at mathandai.org and the Caltech open letter.
Two documents, two days
It is easy to read this week as one protest. It was two, and the order matters.
On 10 September a coalition of current and former Caltech mathematicians published an open letter against the Caltech Mathathon, a hackathon due to start on 30 October in which participants are encouraged to attack open research problems by prompting large language models. The letter did not soften its charge: "To put it bluntly, AI companies are engaging in research misconduct." Its argument is that labs announce AI-generated results for the prestige, then leave human mathematicians to check, explain and sometimes debunk them. "This labor goes uncompensated, uncredited, and unacknowledged," the coalition wrote, and it asked the organisers to cancel the event.
That evening Roberts posted on X: "After considering concerns raised by members of the mathematics community, we've decided to withdraw OpenAI's sponsorship of the Caltech Mathathon and have notified the organizers." We found no similar announcement from Anthropic.
The next day the declaration went up, and Tao reposted it on his blog, writing that he was "proud to be among the list of 25 initial signatories — all Fields Medallists — to the declaration below, which grew out of discussions between ourselves over the last week." Tao is a signatory, not the author. The declaration carries no single author's name, and its endorsement page accepts names only from researchers who verify through ORCID or an academic email address.
What the medallists actually say
The declaration is not a list of demands. It argues that famous problems were always a proxy for the real goal, "conceptual understanding and insight," and that the rapid mass production of true-or-false results "could destroy fertile ground instead of breathing life into new ideas." Its sharpest passage is about how results are released: "Often these solutions are announced in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work of others. As in all creative professions, this raises severe attribution and plagiarism questions."
It names no company, model or result, and its only call to action is that the issues "must be addressed urgently" by mathematicians, by the companies and by society at large. The context is easy to supply. Earlier this week OpenAI said an internal model had resolved the Navier–Stokes Millennium Prize problem. The announcement was followed by a credit dispute, after NYU mathematician Tristan Buckmaster said OpenAI asked him to drop an Anthropic co-author, and by a data question, after OpenAI wrote that it cannot rule out that de-identified user chats helped improve the model.
Why a hackathon drew the fire
A hackathon rewards speed. The Caltech letter's worry is the week after. Picture a restaurant that serves meals at record pace and leaves the washing-up to diners at the next table: results can be announced within a weekend, but verifying a new proof takes experts weeks or months, and nobody pays them to do it. The letter notes that the rules did not require participants to post their results publicly on arXiv, where other mathematicians could scrutinise them.
The organisers answered the same day. They said they are "now making arXiv publication an explicit requirement for the verification period," and aim to finalise speakers, judges and participants by the end of September.
Why it matters
Mathematics has become the AI industry's favourite scoreboard. Problems are famous, answers can be checked, and a solved conjecture makes headlines, a dynamic covered in our explainer on how AI is benchmarked. The declaration is the field's most senior members saying collectively that treating mathematics as a leaderboard spends something the field cannot easily replace, echoing Tao's earlier warning that good problems are a non-renewable resource. And a lab withdrawing a sponsorship within hours of a letter shows that the community has leverage.
There is also a constructive counterexample from the same week: NVIDIA researchers published the complete recipe, data and proofs behind an olympiad gold-medal score, which is the kind of full write-up the declaration says rushed announcements skip.
The strongest objection
In the Hacker News discussion, the most common counter-argument was economic: that this is a predictable reaction to threatened livelihoods and status, expressed in academic language. A more careful version held that AI has not destroyed mathematical understanding but the yardstick, solving open problems, that the field used to hand out credit. Others pointed out that the declaration contains no concrete ask. Under the Caltech letter, commenter Yongxi Lin argued that "slop" should mean incorrect or unreadable work whoever produced it, and opposed cancelling the event.
The honest caveats: the declaration carries no date on its own page (the date comes from Tao's post), endorsement counts are changing by the hour, and OpenAI has not responded to the declaration itself.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)