Summary
On September 11, 2026, twenty-five recipients of the Fields Medal published a joint declaration, "A Severe Misalignment of AI in Mathematics," stating that "the push by AI companies to solve mathematical problems as a benchmark is detrimental to the science of mathematics" and that the goals of the companies and of the mathematical community "are severely misaligned." The statement, whose signatories range from Pierre Deligne (1978) to Yu Deng (2026), was released on a dedicated website, on Terence Tao's blog and in a French version in Le Monde, three days after OpenAI announced that an internal model had resolved the Navier-Stokes Millennium Prize problem. It argues that solving problems is "only a tool and proxy" for conceptual understanding, that rushed announcements raise "severe attribution and plagiarism questions," and that the issues must be addressed "urgently" by mathematicians, the companies and society. It proposes no specific mechanism and does not call for a halt to AI use in mathematics.
What Happened
The declaration appeared on September 11 at mathandai.org, with a Zenodo DOI (10.5281/zenodo.22737750), and on Tao's blog at 17:27 UTC, followed by a Mastodon post from Tao at 17:40 UTC. Tao wrote that he was "proud to be among the list of 25 initial signatories," that the text "grew out of discussions between ourselves over the last week," and that further signatures were invited "similarly to the Leiden declaration." He added: "It is unfortunate that we did not have the time to have a more consultative process, as with Leiden; but we decided that the urgency of the situation was such that we needed to release a statement sooner rather than later." The Economist published an article on the declaration the same day, and Le Monde carried a French version. The website invites endorsements from anyone who signs in with an ORCID identifier or confirms an academic email address.
The twenty-five initial signatories are Artur Avila, Manjul Bhargava, Caucher Birkar, Pierre Deligne, Yu Deng, Simon Donaldson, Hugo Duminil-Copin, Alessio Figalli, Martin Hairer, June Huh, Maxim Kontsevich, Elon Lindenstrauss, Pierre-Louis Lions, James Maynard, Curtis McMullen, Shigefumi Mori, Ngô Bảo Châu, Andrei Okounkov, Peter Scholze, Stanislav Smirnov, Terence Tao, Maryna Viazovska, Cédric Villani, Wendelin Werner and Efim Zelmanov. The Fields Medal is awarded every four years to at most four mathematicians under forty; the list spans eleven award cycles and includes one of the 2026 medallists, Yu Deng.
The text runs to roughly 750 words and makes an argument rather than a demand. It opens by acknowledging that "the mathematical capabilities of LLMs have improved dramatically, to the point that they can solve major outstanding problems in many fields of mathematics." It describes famous problems as "landmarks and lighthouses" whose value lies in the "long and arduous process of talks, discussions, simplifications" that follows a solution and ends, ideally, in "a textbook presentation." "Solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight," it states; "the mass production at faster and faster pace of 'true/false' statements could destroy fertile ground instead of breathing life into new ideas." On announcements, it says: "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 frames the situation as "a general threat to intellectual work" that other scientific and creative professions face, concedes that "AI offers the potential of enhancing and accelerating genuine mathematical study and understanding," and closes by saying that whether the changes "benefit the field or have a destructive effect will in large part be determined by the decisions of the humans in control of this new technology."
The immediate trigger was OpenAI's September 8 announcement that an unreleased internal model, described as "significantly more capable than GPT-6 Astra," had produced a proof that the three-dimensional Navier-Stokes equations can develop a finite-time singularity from smooth initial data with a smooth applied force, resolving one of the seven Millennium Prize problems. OpenAI said it launched the effort on September 1 after hearing "rumors that two Millennium Prize problems had been resolved," used a system of coordinating agents in which the group that produced the result "involved on the order of 10,000 concurrent agents," obtained the resolution on September 5 about 88 hours after launch, and spent a further 17 hours on a Lean formalization using GPT-6 Astra. The post said the rumor "later" turned out to concern Levent Alpöge, an Anthropic employee, and Tristan Buckmaster of New York University, whose forced Euler blow-up result OpenAI said it recognized as having priority; it stated that "we (the researchers and the agents) did not see any of their work through any means until they released it publicly," while adding that "we cannot rule out that de-identified data derived from their usage of our products helped improve our models." Buckmaster, announcing his own three results with Alpöge on September 8, accused OpenAI of pressuring him not to credit his collaborator and questioned whether the company's route to the solution, which he said resembled their own, was "the direction one arrives at in a few days by giving a model the problem statement," according to TechCrunch and NBC News; OpenAI denied the allegations. TechCrunch reported that on September 10 OpenAI withdrew its sponsorship of a mathematics event at Caltech after criticism from researchers there, and that some mathematicians had begun to worry that their use of Codex was feeding OpenAI's models.
The declaration follows the Leiden Declaration on Artificial Intelligence and Mathematics of June 2, 2026, a community initiative endorsed by the International Mathematical Union that set out recommendations for individuals, institutions, governments and industry after a consultative drafting process. In the days after September 11, Tao's blog carried a series of guest essays on the same theme by Burt Totaro, Claire Voisin, Silvia De Toffoli and Eamon Duede, Steven Strogatz, Nestor Guillen, Bryna Kra and Emily Riehl, and opened a crowdsourced list of writings on the purpose and value of mathematics. As of September 16, Tao's Mastodon post had been boosted 648 times and favourited 637 times.
Why It Matters
The mathematical community has been the clearest case of a scientific field where frontier models moved from competition performance to open problems within about a year, from olympiad gold medals in 2025 to the Erdős unit-distance disproof in May 2026 and the Navier-Stokes claim in September. The Erdős result was verified and written up by nine external mathematicians on the day of the announcement; the Navier-Stokes proof remained unverified by the community when the declaration was published. The signatories are not disputing that the results are correct or that the tools are powerful; their claim is that the value the field produces lies in a slow, human process of digestion and transmission that a benchmark race bypasses, and that a company optimizing for announcements captures the proxy while destroying the thing it stood for. Twenty-five medallists signing within a week, without the consultation the Leiden process used, is a measure of how quickly the community's posture changed after September 8.
What the declaration asks for is left open. It names no rule on attribution, disclosure, embargo or credit, and it does not say what "urgently" addressing the issues would look like for OpenAI, Google DeepMind or Anthropic, whose employee's work is at the center of the priority dispute. Whether the companies respond with anything more concrete than the sponsorship withdrawal, whether journals and prize committees adopt policies on machine-generated proofs, and whether the endorsement list grows beyond the initial twenty-five are not yet known. The declaration's own framing, that mathematics is "a miniature version of humanity" and that other creative professions "will confront similar problems," is the reason this document may be cited beyond mathematics: it is the first statement by the top rank of a research discipline that treats the labs' choice of benchmark as itself a harm.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.