OpenAI Drops Its Caltech Math Sponsorship as 25 Fields Medalists Warn of AI’s ‘Severe Misalignment’

Twenty-five Fields Medal winners signed a declaration this week warning that AI companies are “severely misaligned” with the mathematics community, and OpenAI pulled its sponsorship of an upcoming Caltech math competition after a separate letter from a coalition of mathematicians called the event a driver of “slop mathematics.” Both developments follow OpenAI’s September 8 announcement that an internal model had resolved part of the Navier-Stokes existence and smoothness problem, and NYU mathematician Tristan Buckmaster’s complaint that a parallel OpenAI effort drew on his and a collaborator’s unpublished work to publish a complete proof before they could finish their own.

A Credit Dispute Kicks Off the Week

On September 8, Buckmaster announced three proofs of his own, developed with Anthropic researcher Levent Alpoge using Codex and Claude, including preliminary progress on a related fluid-dynamics problem. In his own statement, as TechCrunch reported, Buckmaster wrote that “there is another part of this story, and one that, honestly, I very much wish I did not have to be concerned with,” describing a parallel OpenAI effort that built on their work before it became public. That same day, OpenAI published its own Navier-Stokes proof, attributed to an unreleased model whose agents reached the result in roughly 88 hours, part of a week-long, multi-problem sweep that OpenAI says consumed 300 billion output tokens in total, which TechCrunch calculated as roughly $22.5 million in compute at current pricing, as sxz.io reported at the time. This week, according to TechCrunch, Buckmaster went further, accusing OpenAI of pressuring him not to credit Alpoge and questioning whether the company had fed their Codex-assisted work back into its own model to produce the announcement.

25 Fields Medalists Sign a Misalignment Declaration

That dispute set off a faster, broader reaction than OpenAI’s proof itself. On September 11, Terence Tao published a joint declaration titled “A Severe Misalignment of AI in Mathematics” on his blog, co-signing it with 24 other Fields Medal winners, mathematics’ most prestigious prize. Tao wrote that the statement “grew out of discussions between ourselves over the last week” and that the group skipped a more deliberate drafting process, similar to the one behind June’s Leiden Declaration on AI and mathematics, because “the urgency of the situation was such that we needed to release a statement sooner rather than later.” Signatories span nearly five decades of the prize’s history, from Pierre Deligne, who won in 1978, to this year’s recipient, Yu Deng.

The declaration itself, hosted at mathandai.org, argues that “the push by AI companies to solve mathematical problems as a benchmark is detrimental to the science of mathematics, and to the mathematical community,” and that rushed announcements leave “no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work of others.” The signatories frame this as “severe attribution and plagiarism questions” and warn that without mathematicians willing to verify and integrate AI-generated results into the field’s body of knowledge, “AI-conceived ideas would never become fully alive.” The letter does not name OpenAI specifically, framing the concern as an industry-wide dynamic, but its timing, the same week as Buckmaster’s renewed accusation, makes the connection hard to miss.

A Separate Letter Targets a Caltech Event

A second, distinct document emerged from a different direction. A coalition of current and former Caltech mathematicians published an open letter on the blog Proofs and Prompts, dated September 10 but received September 8, the same day as Buckmaster’s statement, objecting to the Caltech Mathathon, a hackathon scheduled for October 30 in which undergraduate teams compete to solve open research problems using AI models, backed by $2 million in combined credits from OpenAI and Anthropic. The letter argues the event will produce “slop mathematics” that undergraduates lack the expertise to verify, and states bluntly that “AI companies are engaging in research misconduct” by using the results of unpaid mathematical labor to promote their products without requiring participants to publish verified writeups. As of publication, the letter had drawn 771 signatories, open to anyone at the PhD level or higher in mathematical research or affiliated with the Caltech community.

Unlike the Fields Medal declaration, this letter is specific to one event and does not require Fields Medal status to sign, though its underlying complaint, that AI labs are extracting research value from mathematicians without compensation or credit, echoes the broader declaration published the next day.

OpenAI Pulls Its Sponsorship

OpenAI research lead Dan Roberts announced on X on September 10 that the company would withdraw its Mathathon sponsorship, writing that “we recognize that the rapid progress of AI in mathematics is disruptive” and that OpenAI wants “to engage with the math community more on the best way to integrate this technology and communicate its impacts.” Caltech’s Mathathon organizers, for their part, told Gizmodo they expect the event to proceed largely unaffected: “We do not anticipate that this will affect the event in any substantial way.” Anthropic, the competition’s other named sponsor, had not made a similar announcement as of publication.

None of this resolves the underlying dispute over Buckmaster and Alpoge’s work, and the Clay Mathematics Institute’s own rules mean any formal recognition of OpenAI’s Navier-Stokes claim is still years away regardless of how the credit question shakes out. But the sequence, a contested proof, a rushed declaration from the field’s most decorated researchers, and a sponsor pulling out of a student competition within days of each other, illustrates how quickly frontier labs’ math claims are now generating organized pushback from the community whose work they depend on.