Mathematics has a part to play in making advanced AI safe, and much of that work is still waiting for mathematicians to take it up. MAIS is where that work happens in public: precisely stated open math problems, each paired with a theory of change for how a solution could advance AI safety; research agendas that incorporate multiple open problems and their surrounding context; and papers that grow out of solutions to the problems.
This repo hosts pre-publication math research of three kinds:
- Open Problems — a short exposition of a single open problem
- Agendas — a cluster of related open problems and partial progress on resolving them
- Papers — papers in progress
The intended life cycle is open problem → agenda → paper → publication: open problems get incorporated into agendas, and a maturing agenda spins off papers. Cited external work is summarized in references/, one page per paper, suitable for stuffing an LLM's context.
Each document carries a unique permanent identifier — MAIS-On for open problems, MAIS-An for agendas, MAIS-Pn for papers — so a reference like MAIS-O1 or MAIS-A4 stays valid even if the title changes.
The open-problem index is the full registry, and the agenda guide offers a first foothold in each research direction. Conventions for numbering, structure, and building live in HOUSEKEEPING.md.
Contributions of every size are welcome; see CONTRIBUTING.md.
The best first move is to open an issue — a public thread that anyone can answer. A substantial contribution earns an invitation to join the author list, subject to the judgment of the document's current authors.
MAIS grew out of Lionel Levine's survey paper Math for AI Safety: An Invitation for Mathematicians. The repository is maintained by Lionel together with
Maintainers review contributions, verify claimed results, and merge pull requests; see CONTRIBUTING.md for how to work with them.
Cite documents by their permanent identifier, as their authors' own work. Templates in CITATION.md.
All text and sources are released under CC BY 4.0.
- Math for AI Safety: An Invitation for Mathematicians — a survey organized by mathematical field, each section ending with an accessible open problem.
- Where to start, as a mathematician — a thread on X: what AI safety is, how to learn more, and where math helps, with links in reading order.
- AI-Safety for Mathematicians — a starting point for professional mathematicians who want to engage with AI safety, maintained by Jacob Tsimerman.
- Statistics for AI Safety — a collection of self-contained open problems in AI safety for statisticians.
- The Alignment Project — the research agenda of a UK AISI–led coalition funding alignment research; several of its priority areas are mathematical.
- AI + Mathematics Resources — a curated list from Northwestern's math department, including research platforms, formalization projects, educational programs, funding, conferences, and mathematicians' perspectives.