Particle.news

Leiden Declaration Seeks New Rules for AI Use in Mathematical Research

The document sets norms for disclosure, stronger peer review, public computing investment, protections for verification and attribution.

Overview

  • The Leiden Declaration was published Tuesday, June 2, 2026, by 16 authors and had gathered more than 130 signatories calling for community standards on AI in math.
  • The statement grew from a Lorentz Center workshop last fall where about 60 researchers and policy makers drafted an 11‑page text after months of debate.
  • Authors warn that AI‑produced proofs can contain subtle errors that are hard to verify and that commercial labs sometimes delay or withhold methods, citing examples such as DeepMind’s AlphaProof episode.
  • Key recommendations include mandatory disclosure of AI use in papers, tougher peer review for AI‑derived results, clear attribution practices, and public investment in computational infrastructure.
  • The International Mathematical Union has engaged with the declaration and plans formal endorsement and discussion at its upcoming Philadelphia meetings, a step that could prompt journal, institutional or policy changes and affect how researchers publish and verify results.