Overview
- A leading AI lab announced that its models solved one of mathematics’ best‑known problems and released machine‑checkable files while saying it is withholding many other results.
- The proofs produced by AI can be formally verified in systems like Lean yet are often opaque to human readers because they skip key steps, fail to explain ideas, and do not relate results to existing work.
- Mathematicians and students have reacted with anger and worry, prompting public confrontations, the creation of advocacy groups, and behavioral shifts such as posting unfinished work earlier or removing open conjectures to avoid scraping.
- Scholarly infrastructure is straining: arXiv has limited submissions, journal editors report unresponsive or unverifiable AI‑assisted papers, and companies face growing disputes over training data and credit.
- The field now faces two paths: a risky erosion of incentives for human theorem proving that could weaken training and careers, or an adapted role for mathematicians focused on interpreting, organizing, and building research programs around AI outputs.