Jacob Tsimerman, winner of the 2026 Fields Medal, announced the founding of the Mathematical AI Safety Institute (MAISI) on Sept. 8, an independent nonprofit the group describes as "developing mathematical foundations for the safety of powerful AI systems." Andrew Critch -- a mathematician and former researcher at Berkeley's Center for Human-Compatible AI, and Tsimerman's co-author on a 2025 paper about catastrophic AI scenarios -- will serve as executive director. The first cohort targets 10-30 mathematicians for a January 2027 semester, scaling to 30-100 researchers for a more ambitious "Special Year" in September 2027. The pitch, in the institute's own words: "The foundations of statistics, physics, and even computer science are mathematical... Now, we need similar mathematical foundations for AI safety, not only to measure risk, but to mitigate it." The stated problem, per the institute's site: "we lack a rigorous understanding of what it would mean to be safe, even in theory" -- a gap MAISI wants to close with formal definitions and measurements, not additional red-teaming or policy work, which is where most AI safety funding currently goes.
The launch, in short
- Scientific Director
- Jacob Tsimerman
- Executive Director
- Andrew Critch
- Model
- Princeton's Institute for Advanced Study
- First cohort
- 10-30 mathematicians
- Scale-up
- 30-100 researchers
Independence is the institute's own framing, and it sits next to a real tension: Tsimerman is simultaneously joining OpenAI's safety research team this month, even as MAISI describes itself as operating independently of any lab. The institute's advisory panel leans toward names already inside frontier labs or their orbit -- Geoffrey Irving, chief scientist at Resolution and a former DeepMind and OpenAI alignment lead, and Paul Christiano, director of the Alignment Research Center -- alongside Stanford mathematician Ravi Vakil and 1998 Fields Medalist Timothy Gowers. MAISI's own materials do not disclose a funding source.
The launch lands the same week a dispute over credit for an AI-assisted proof of a Navier-Stokes blowup problem was dominating mathematicians' attention online -- a coincidence of timing, not a connection either announcement makes, but a reminder that the same week AI's mathematical capabilities and its safety-research funding both made news through mathematicians rather than labs.
- Fields Medalist Jacob Tsimerman launched the Mathematical AI Safety Institute (MAISI) on Sept. 8.
- Andrew Critch, his 2025 co-author on catastrophic-AI-risk scenarios, is executive director.
- MAISI aims to hire 10-30 mathematicians for a Bay Area semester starting January 2027.
- The pitch is rigorous mathematical definitions for AI safety, modeled on Princeton's IAS.
- Tsimerman is simultaneously joining OpenAI's safety team -- the institute says it stays independent.