The International Mathematical Union has endorsed the Leiden Declaration on Artificial Intelligence and Mathematics, a document published June 2, 2026 that warns AI tools and tech-industry incentives are eroding the standards of mathematical research. The declaration was drafted by a working group of 16 researchers over 8 months and has already attracted hundreds of signatories, including Fields Medal-affiliated mathematicians. It arrives two weeks after OpenAI publicized an AI model that it said disproved an 80-year-old conjecture in geometry.
The timing is not incidental. The declaration explicitly criticizes corporate press releases that announce mathematical achievements on "market timelines before the accepted processes of community evaluation in mathematics can take place." OpenAI's announcement coincided with reporting that the company was preparing to offer stock to the general public, a juxtaposition the declaration's authors point to as the pattern they want to push back on.
The working group convened after a September 2025 conference at Leiden University in the Netherlands. Kevin Buzzard, a mathematician at Imperial College London, framed the document as a measured response rather than an outright rejection of AI in research.
Key facts
- 01The Leiden Declaration on Artificial Intelligence and Mathematics published June 2, 2026 with endorsement from the International Mathematical Union.
- 0216 researchers developed the declaration over 8 months following a September 2025 conference at Leiden University.
- 03The document landed two weeks after OpenAI publicized an AI model that disproved an 80-year-old geometry conjecture.
- 04Hundreds of signatories have endorsed the declaration, including Fields Medal-affiliated mathematicians and the IMU vice president.
- 05Signatories include Peter Scholze of the Max Planck Institute and Kevin Buzzard of Imperial College London.
The declaration lays out five concerns. AI models can produce plausible but unreliable arguments that are difficult to distinguish from correct proofs, which the authors say is "jeopardizing our ability to implement traditional standards for the correctness, transparency, and independent verifiability of proof." Leslie Ann Goldberg, head of computer science at the University of Oxford, warned that cheap AI drafts risk "cluttering the literature with claimed results that are simply wrong" — errors that then propagate as later work builds on them.
The document also flags citation gaps in model outputs, training data obtained by "exploiting licenses and access arrangements" or violating copyright, and the risk that AI use becomes incentivized for its own sake in hiring, funding, and recognition. A fifth concern is the growing involvement of tech companies in mathematical research at a moment when university budgets are tight, which the authors argue pushes academics into "asymmetric terms" of collaboration.
Rodrigo Ochigame, a historian and anthropologist of computing at Leiden University and one of the declaration's authors, pointed to OpenAI's recent announcement as a specific example. OpenAI uploaded a paper describing the model's proof along with outside commentary, but did not disclose the prompts used, training data, or compute resources consumed.
Michael Harris, a mathematician at Columbia University and an author of the declaration, told The New York Times that "the tech industry proceeds in accordance with commercial logic, which is antithetical to the values of mathematics," and described the declaration as an attempt to "recover control of the narrative" about mathematics' goals from the AI industry. Ursula Martin of Oxford University called the OpenAI result "remarkable" but noted that equivalent human effort would likely have produced the same proof, and that mathematics is also about "cultivation of ideas, understanding, judgment and human insight."
Peter Scholze, director of the Max Planck Institute for Mathematics and one of the more prominent endorsers, framed his objection in personal terms.
The declaration's recommendations are concrete. Individual mathematicians are asked to disclose AI use, retain responsibility for correctness, properly credit human authors, and consider whether the tools they use align with the declaration's values. Professional organizations are asked to develop AI guidelines for publication and review, protect researchers' work from being used as training data without consent, and "actively prepare to become involved if major mathematical results are claimed using unconventional means." For policymakers, the asks are to protect author rights, regulate the AI industry, and invest in public computational infrastructure.
The document also acknowledges the pull factor working against it: the tech industry "has offered lucrative jobs, monetary rewards, computing resources, and intellectually stimulating opportunities" in an era of underfunded higher education and precarious academic employment. It does not tell mathematicians to refuse those offers — it asks that any such collaborations abide by the standards in the declaration.
Ulrike Tillmann, vice president of the International Mathematical Union, said the IMU's endorsement 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." Whether that affirmation slows the integration of AI tools into peer-reviewed math is another question — Buzzard himself has been a longtime advocate of formal proof systems and has worked with AI-adjacent tools, suggesting the signatories are not opposed to automation so much as to its commercial framing.
The Leiden Declaration is the clearest sign yet that an established scientific discipline is trying to set terms with the AI industry before the industry sets terms with it. For OpenAI, Google DeepMind, and other labs that have been chasing mathematical benchmarks as proof points for general reasoning, the document is a reminder that the people who actually run the field are now organized, named, and on the record. The next test will be whether the IMU's affiliated journals adopt the disclosure and review standards the declaration outlines — that, more than the signatures, is what would change what gets published and how AI-assisted results are evaluated.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
