Overview
- The Leiden Declaration, published Tuesday, was written by a 16‑person working group, has drawn more than 100 signatories and received endorsement from the International Mathematical Union.
- Authors warn that large language models can produce plausible but unreliable proofs that are hard for human reviewers to check, which risks cluttering the literature with incorrect results.
- The declaration criticises tech firms for promoting mathematical claims on market timelines while withholding prompts, training data and compute details that are needed for independent evaluation.
- It lays out concrete steps for individuals, professional bodies and policymakers, including mandatory disclosure of AI use, stricter peer review, protections for authors over training data, public investment in compute infrastructure, and regulatory attention to industry practices.
- The statement is prompting talks within major mathematical organizations and will shape guidance at upcoming meetings in Philadelphia while raising concerns about access gaps that could disadvantage students and early‑career researchers.