More than 150 mathematicians signed the 11-page Leiden Declaration urging governments not to accept flashy AI proof claims without expert review. The declaration warns that commercial incentives can lead companies to overstate AI capabilities and that unvetted AI-generated proofs could misdirect research and erode trust. It also flags risks—misinformation, surveillance, military use and environmental costs—while acknowledging AI’s practical benefits for energy systems and other applications.
150+ Mathematicians Urge Caution: Don’t Take Headline AI ‘Proofs’ at Face Value
Photo Credit: iStock
More than 150 mathematicians and related scholars have signed the 11-page Leiden Declaration on AI and Mathematics, warning governments and policymakers not to accept dramatic claims about AI-generated mathematical proofs without careful expert review.
Why the Declaration Was Issued
The declaration responds to a series of high-profile incidents this year that drew public attention to AI and mathematics. Among them: a widely publicized claim by a 23-year-old that ChatGPT helped solve one of several well-known Erdős problems, and an OpenAI statement asserting its system had refuted the decades-old "unit distance" conjecture attributed to Paul Erdős. Many researchers say such announcements fall well short of demonstrating that AI has reliably transformed mathematical research.
Key Concerns
"There is currently a strong commercial incentive on the part of the technology industry to overstate the capabilities of their products,"the declaration states, urging policymakers not to "believe the hype." Signatories emphasize that mathematics underpins science, engineering, economics, computing and climate modeling; flawed or unvetted results could misdirect research, delay real breakthroughs and erode trust in the scientific record.
The declaration also highlights broader risks tied to rapidly scaled AI: misinformation, intrusive surveillance, military applications and substantial environmental costs. Training and operating advanced models often requires vast data centers that place pressure on energy grids, consume large amounts of water and contribute to pollution and infrastructure stress.
Potential Benefits—and the Need for Balance
The statement does not dismiss AI outright. It acknowledges practical, beneficial uses—for example, helping utilities forecast electricity demand, integrate renewable resources and optimize grid efficiency. The central message is that benefits should not be conflated with definitive claims about foundational scientific advances without rigorous verification.
Voices From the Field
Ulrike Tillmann, vice president of the International Mathematical Union, said AI "raises questions that cannot be left unexamined." Leslie Ann Goldberg, head of computer science at the University of Oxford, warned: "Current automated techniques can produce plausible but unreliable (or even incorrect) arguments which are difficult to distinguish from correct mathematical proofs." Rodrigo Ochigame, an anthropologist of AI at Leiden University and a contributor to the declaration, expressed concern that mathematicians' work is being repurposed for AI development without their consent.
Recommendations for Policymakers
The Leiden Declaration calls on governments to consult with domain experts—professional mathematicians and researchers—when evaluating claims about AI-driven mathematical breakthroughs. It recommends skepticism of press releases and media summaries that present preliminary or unverified results as conclusive, and it urges assessment of broader societal impacts when shaping AI policy.
Bottom line: Exciting AI headlines warrant attention, but they also require careful, expert scrutiny before policymakers treat them as proof of sweeping scientific progress.
Help us improve.
