News & resources
News on artificial intelligence and mathematics: what is happening, in brief, and where to read more. Collected by WINDSMATH.
Report Oct 2026
The Martingale Foundation surveyed masters and PhD students, and academics and supervisors, in October 2026, collecting more than 260 responses. Its briefing note looks at how rapid developments in AI are changing mathematical research, weighs the threats against the opportunities, and asks what support would help postgraduates most.
Summit Sep 2026
At a two-day summit, a group of mathematicians set out recommendations for PhD programmes: new training in collaboration, communication, AI tools and formalisation; assessment based on in-person written or oral exams; access to AI tools for every graduate student, but never mandatory; and three ethical rules: take responsibility for the mathematics under your name, always disclose AI use, and get agreement before entering anyone else’s material into an AI system.
Essay Aug 2026
In an essay adapted from his public lecture at the 2026 International Congress of Mathematicians, Terence Tao sets aside whether AI tools will become capable of research-level problems. He asks instead how the community should prepare, and what mathematical research is fundamentally for, using problem-solving as a case study.
Feature Jun 2026
As AI produces proofs and research results, mathematicians are reconsidering the purpose and future of their discipline. IEEE Spectrum gathers views ranging from AI as a simple tool to AI as a collaborator, alongside concerns about unequal access, weakened incentives and students developing fewer skills of their own.
Feature Apr 2026
After strong results at the 2025 International Mathematical Olympiad, AI models began helping mathematicians produce new research results, from the proof of a decades-old optimisation conjecture to newly discovered structures in permutation groups. Quanta also reports the worries: low-quality AI-generated submissions reaching journals, and students relying on AI instead of learning to solve problems.
Essay Mar 2026
AI systems can now prove advanced theorems, both in formal proof systems and in ordinary informal reasoning. Jeremy Avigad urges mathematicians to keep pace with these tools, anticipate how they will reshape mathematical work, and adapt to the challenges and opportunities that follow.
Paper Aug 2025
Surveying publicly available language models, Jonas Henkel finds the best of them strong at solving problems and evaluating proofs, but prone to skipping checks on their own reasoning and to reaching correct answers without fully valid proofs. He proposes a human-led workflow, five principles across seven stages of research, in which AI augments the mathematician.
Tools
Lean is a proof assistant, and Mathlib is a community-driven library of mathematics formalised in it. For anyone who wants to try formal proof, the Lean community site has tutorials, books and an online editor that needs no installation.
