You know that unsettling feeling when ChatGPT solves a complex problem but you can’t follow its reasoning? Mathematicians are experiencing something similar—except the stakes involve the foundation of scientific certainty itself. A new document called the Leiden Declaration on Artificial Intelligence and Mathematics argues that AI’s rapid deployment in mathematical research threatens to erode the discipline’s core values: rigorous proof, transparent reasoning, and human understanding.
The Academic Equivalent of a Warning Shot
Sixteen mathematicians drafted a consensus document that now has backing from over 130 researchers and the International Mathematical Union.
The declaration emerged from a September workshop in the Netherlands, where mathematicians spent months hammering out every detail until they reached full consensus. “We did this the hard way,” explains Rodrigo Ochigame, an anthropologist of AI who participated in the process. The International Mathematical Union’s endorsement signals this isn’t fringe anxiety—it’s mainstream concern from the mathematical establishment.
The Blackbox Problem Hits Pure Math
AI systems can now generate mathematical proofs that humans struggle to verify or understand.
Here’s where things get unsettling: frontier AI models are producing increasingly sophisticated mathematical arguments that peer reviewers can’t easily check. Unlike a calculator showing its work, these systems output complex proofs from internal processes we don’t fully understand. Daniel Litt from the University of Toronto warns of a “rush to announce results” from AI startups whose findings are “mostly correct and also not very interesting”—but whose marketing suggests otherwise.
The attribution problem runs deeper. AI systems train on arXiv, the open repository where mathematicians share preprints, then generate outputs without clear citations. It’s like having a brilliant student who absorbed everyone’s homework but can’t explain which ideas came from where.
Proposed Remedies Target the Entire Pipeline
Mathematicians want mandatory AI transparency and publicly funded alternatives to corporate tools.
The declaration’s solutions are surprisingly concrete:
- Mandatory disclosure of AI use in research
- Stricter peer review that can handle AI-assisted work
- Investment in public computational resources to counter Big Tech‘s growing influence over mathematical discovery
“Mathematics is, and should always remain, a profoundly human endeavor,” wrote IMU vice president Ulrike Tillmann. The sentiment captures the declaration’s central tension: embracing AI’s power while preserving human insight and community trust.
This isn’t mathematical Luddism—it’s a recognition that when the foundations of scientific certainty become opaque, everything built on top becomes shakier. Your smartphone‘s encryption, your GPS navigation, the algorithms that run modern finance—they all rest on mathematical proofs. If mathematicians can’t trust their own field’s integrity, that uncertainty propagates everywhere.
