AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Source Summaries

A Severe Misalignment of AI in Mathematics (Fields Medallists' declaration, 2026)

high confidence · updated 2026-09-13

Declaration signed by 25 Fields Medallists arguing that the use of mathematical problem-solving as an AI benchmark damages mathematics as a discipline, and framing the divergence between AI companies' goals and the profession's goals as a misalignment problem extending to other intellectual work.

Contents

"A Severe Misalignment of AI in Mathematics" is a collective declaration signed by twenty-five Fields Medallists and published at mathandai.org. It argues that the practice of AI developers using mathematical problem-solving as a capability benchmark is detrimental to mathematics as a science and to the mathematical community, and that the goals of AI companies and of the mathematical community are "severely misaligned." The signatories present the mathematical case as one instance of a wider divergence affecting "other scientific and creative professions, as well as the whole of society." The declaration carries no date on its own face; 11 September 2026 is the date of signatory Terence Tao's blog post of the same title and of the surrounding coverage.

Signatories

The signature block carries twenty-five names with their Fields Medal years: Artur Avila (2014), Manjul Bhargava (2014), Caucher Birkar (2018), Pierre Deligne (1978), Yu Deng (2026), Simon Donaldson (1986), Hugo Duminil-Copin (2022), Alessio Figalli (2018), Martin Hairer (2014), June Huh (2022), Maxim Kontsevich (1998), Elon Lindenstrauss (2010), Pierre-Louis Lions (1994), James Maynard (2022), Curtis McMullen (1998), Shigefumi Mori (1990), Ngô Bảo Châu (2010), Andrei Okounkov (2006), Peter Scholze (2018), Stanislav Smirnov (2010), Terence Tao (2006), Maryna Viazovska (2022), Cédric Villani (2010), Wendelin Werner (2006), and Efim Zelmanov (1994). The medal years span 1978 to 2026.

Endorsement is a separate mechanism from signature. The page invites readers to endorse the declaration through a distinct form, with endorsements collected at mathandai.org/sign.php and listed on a separate endorsers page; the site states that an endorsement appears after the endorser signs in with ORCID or confirms an academic email address. The endorser list is not an extension of the twenty-five-name signatory block.

Summary of the argument

The declaration opens by conceding the capability it goes on to criticise: over the preceding months the mathematical capabilities of large language models have "improved dramatically, to the point that they can solve major outstanding problems in many fields of mathematics." The objection is to the use made of that capability rather than to the capability itself.

The central distinction is between solving a problem and understanding a subject. The declaration describes famous open problems as having served historically as "landmarks and lighthouses against which one can measure an improved understanding" of the mathematical landscape, with a solution functioning as evidence of new insight and method. That insight then passes through what the signatories describe as a long process of talks, discussions and simplifications, ending ideally in a textbook presentation accessible to a graduate or undergraduate student, and in some cases becoming general public knowledge decades or centuries later. On this account, solving problems is "only a tool and proxy for achieving the primary goal of conceptual understanding and insight," and treating the proxy as the objective "may turn the tool against the primary goal." The declaration states that "the mass production at faster and faster pace of 'true/false' statements could destroy fertile ground instead of breathing life into new ideas."

A second line concerns attribution. The declaration states that machine-produced solutions are "often 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," and that this "raises severe attribution and plagiarism questions" as it does in all creative professions. It adds that without mathematicians willing to develop and integrate them, AI-conceived ideas "would never become fully alive" and the human transmission chain between mathematicians would be lost.

A third line generalises the claim beyond mathematics. The signatories describe "a general threat to intellectual work, with misalignment between the outcome of the use of AI and its initial purpose," arguing that years of training in many fields have produced not only final answers but the capacity to form new questions, and that AI systems able to produce the results of such work directly cause those two goals to cease to align. The declaration frames the question facing the mathematical community as one "that all of humanity might face: how to make sure that, as AI changes the way work is done, we do not lose sight of what that work was meant to achieve in the first place."

The declaration also describes the mathematical community in institutional terms, as functioning "in many ways, as a miniature version of humanity" — individuals using varied approaches joined by shared values, whose "most precious resources" are students and ideas.

Key claims

  • The push by AI companies to solve mathematical problems as a benchmark is detrimental to the science of mathematics and to the mathematical community.
  • The goals of AI companies and the goals of the mathematical community are severely misaligned, and this is an instance of "broader alignment issues" affecting other scientific and creative professions and society generally.
  • Problem-solving is a proxy for conceptual understanding, not the goal; optimising the proxy can work against the goal.
  • Rushed announcement of machine-produced results forecloses proper writeup, isolation of methods, and citation of prior work, raising attribution and plagiarism questions.
  • Human uptake is a precondition for a mathematical idea entering the canon; without it, AI-conceived ideas do not become "fully alive."
  • AI "offers the potential of enhancing and accelerating genuine mathematical study and understanding," and whether the changes benefit or damage the field "will in large part be determined by the decisions of the humans in control of this new technology."
  • The issues must be addressed urgently by the mathematical community, by the companies developing the technologies, and by society.

What the declaration does not say

The declaration contains no call for regulation, prohibition, moratorium or pause. Its operative ask is that the issues "be addressed urgently" by three named constituencies, without specifying an instrument. It states in terms that AI offers potential benefit to mathematical study, and locates the determining factor in the decisions of those controlling the technology rather than in the technology's capabilities. An account presenting the declaration as opposition to the use of AI in mathematics would reverse its stated position.

The word "misalignment" in the title is the signatories' own, and is applied to a divergence between the goals of AI companies and the goals of a profession. That is a different referent from the use of the term on AI Alignment, which concerns the correspondence between a model's behaviour and its designers' intentions.

Context

The declaration follows a sequence of claimed machine-produced mathematical results through 2026, recorded on AI for Science and Machine-produced mathematics: claims, verification and credit: the Erdős unit-distance counterexample in May, the Jacobian conjecture counterexample in July, OpenAI's ten results in mathematics and theoretical computer science in August (Ten Advances in Mathematics and Theoretical Computer Science), and the Navier–Stokes finite-time singularity claim of September 8, 2026 (On the Navier–Stokes Millennium Prize Problem (OpenAI, September 2026)). The last of these carried a credit dispute involving Anthropic mathematician Levent Alpöge and NYU professor Tristan Buckmaster, reported three days before the declaration appeared. The declaration itself names no company, model, result or individual.

Signatory Terence Tao published a blog post under the same title on 11 September 2026. In it, according to the 2026-09-11 evening developments digest, he recorded that the signatories forwent a more consultative drafting process of the kind used for the Leiden declaration. That comparison is Tao's and does not appear in the declaration text.

Provenance

Captured from mathandai.org on 2026-09-12 with the full declaration text and the complete twenty-five-name signature block, enumerated rather than inferred from the coverage. The site carries a correction and removal contact address. No DOI or other formal identifier exists for a declaration of this kind. Seven divergences between the declaration and the digest that first carried it are recorded in the raw file's ## Source-fidelity notes; four are material — the distinction between signatories and endorsers, the declaration's scope beyond mathematics, the absence of any regulatory ask, and the absence of a date on the document's face.

Relationships

Sources

  • Raw source: Raw Sources/Declaration of 25 Fields Medallists - A Severe Misalignment of AI in Mathematics (Math and AI, 2026-09-11).md
  • Declaration: mathandai.org
  • Terence Tao, "A Severe Misalignment of AI in Mathematics," terrytao.wordpress.com, 11 September 2026
Categories: Source Summaries | global | opinion | frontier-models | ai-for-science | alignment | academic | safety
What links here (11)