← Back to feed

Sixteen researchers from fifteen universities published the Leiden Declaration on Artificial Intelligence and Mathematics on June 2, 2026, backed by the International Mathematical Union and more than 130 signatories. The document does not call for banning AI from mathematics. It calls for the mathematical community to govern it before market incentives govern the discipline instead.

The timing is not coincidental. A week earlier, OpenAI announced its reasoning model had resolved a 1946 geometry problem first posed by Paul Erdős — a result communicated through a blog post, not peer review. For many mathematicians, that announcement crystallised a pattern that had been building for years.

Five Threats

The declaration names five specific ways AI is undermining what makes mathematics trustworthy.

Unreliable proofs. AI systems produce plausible-looking arguments that are difficult to distinguish from correct proofs. The problem applies to informal reasoning and to formal computer-verified proofs, where the challenge lies in translating between machine and human representations of the same concepts. In a discipline built on certainty, the declaration calls this existential.

Attribution and copyright. Models trained on published mathematical work do not cite the human contributions they synthesise. The declaration states that training data was obtained by “systematically exploiting licences and access arrangements that were not made with artificial intelligence in mind, or indeed by simply violating copyright protections.”

Distorted incentives. AI use is becoming incentivised for its own sake, reshaping hiring decisions, grant criteria, and recognition in ways that do not track intellectual contribution.

Press-release science. Results are increasingly announced through blogs and press releases timed to market cycles, ahead of peer review. The phrase from the declaration: results communicated “on market timelines before the accepted processes of community evaluation in mathematics can occur.”

Lost autonomy. Mathematics risks ceding control of its own research agenda to the commercial priorities of AI companies.

Institutional Weight

Peter Scholze, director of the Max Planck Institute for Mathematics and recipient of the 2018 Fields Medal, endorsed the declaration: “The goal of mathematical research is human understanding of mathematics, so mathematics can only thrive in a community of human researchers. It is crucial to preserve this communal spirit.”

Ulrike Tillmann, vice president of the IMU, wrote that endorsing the declaration “affirms that the future of mathematical research must be guided by human judgment, fair and transparent practices, and the shared values of the global mathematical community.” Tillmann also serves as director of the Isaac Newton Institute for Mathematical Sciences.

IMU Committee on Publishing chair Ilka Agricola was blunter: AI in mathematics “offers fantastic possibilities when used honestly and competently as a research assistant. But this seems to be only a small part of what is going on: the by far larger part is a total mess where science itself is under attack.”

23 Recommendations

The declaration splits its 23 recommendations across three groups.

For individual mathematicians: disclose which AI tools were used and for which purposes; retain human responsibility for correctness; maintain peer review obligations; and “affirm the humanity” of the discipline.

For mathematical organisations and funders: invest in public AI research infrastructure not dependent on proprietary models; develop community standards for attribution; and provide legal resources to help academic institutions defend against training-data exploitation.

For policymakers: four recommendations including government investment in open AI research and legislative action on copyright.

Context

The declaration emerged from a workshop at Leiden University’s Lorentz Center in late 2025, where approximately sixty researchers and policymakers convened to assess AI’s impact on mathematics. Co-organiser Rodrigo Ochigame: “We worked on the declaration for several months, bringing together different points of view in search of shared principles.”

Journal editors are experiencing the practical consequences directly. Inboxes are filling with AI-generated proof submissions faster than reviewers can evaluate them. The Leiden Declaration is the first coordinated response from an international body to that backlog — and to the broader question of whether AI companies should set the pace of mathematical discovery.