An independent non-profit called the Mathematical AI Safety Institute (MAISI) launched on September 8, 2026 under 2026 Fields Medalist Jacob Tsimerman as Scientific Director and Andrew Critch as Executive Director, hiring 10-30 mathematicians to relocate to the Bay Area for a January 2027 semester and 30-100 for a September 2027 Special Year.
An IAS For AI Safety
MAISI is modeled on traditional mathematical research institutes like Princeton's Institute for Advanced Study and Toronto's Fields Institute, hosting Members for one or two semesters at a time with the possibility of renewal. Its stated goal is to develop mathematical foundations — definitions, measurements and solution concepts — for the safety of powerful AI systems, on par with the mathematics physicists use for nuclear reactor safety or statisticians use to track pandemics. The Institute operates independently from any AI lab, though Tsimerman is simultaneously joining OpenAI's safety department in September.
A Panel Of Fields Medals And Alignment Names
MAISI's Scientific Advisory Panel includes Ravi Vakil, Stanford professor and 2025 president of the American Mathematical Society; 1998 Fields Medalist and Chaire de Combinatoire Timothy Gowers; Geoffrey Irving, co-founder and Chief Scientist of Resolution and former Chief Scientist of the UK AI Security Institute; and Paul Christiano, Director of the Alignment Research Center and former head of safety at the US AI Safety Institute. Critch, who holds a 2013 UC Berkeley PhD applying algebraic geometry to machine learning, co-authored a 2025 paper with Tsimerman on "omnicidal" AI risk scenarios.
Three Open Questions MAISI Is Willing To Prove Unsolvable
MAISI's launch statement lists three possibilities its mathematicians will pursue on AI x-risk: that the current AI development paradigm cannot be made existentially safe and needs a paradigm shift; that safe configurations exist but are provably chaotic and sensitive to noise; or that verifying safety has a provably high computational complexity class. Mathematics, the Institute notes, is one of the only disciplines that can prove one of its own problems unsolvable — and if a given AI technology can never be made safe, MAISI wants that fact known before it is too late.
Timing And Structure
Unlike academic mathematics, MAISI plans to circulate ideas in living documents long before they are finalized in journal-ready form. The January 2027 cohort will run alongside partner institutions in AI safety, with a more intensive September 2027 Special Year scaling up to 100 researchers. Applications are open at maisi.org/apply. Related coverage: Anthropic's compute roadmap and OpenAI's GPT-6 Astra Navier-Stokes Lean proof.
Reporting based on coverage from maisi.org, The Hill, NewsNation and Cointelegraph.