Mathematical AI Safety Institute launches to build mathematical foundations for AI safety

Summary

The Mathematical AI Safety Institute (MAISI) is an independent nonprofit focused on developing mathematical foundations for the safety of powerful AI systems. It is led by mathematicians and AI safety researchers, including Scientific Director Jacob Tsimerman, Executive Director Andrew Critch, and an advisory panel that includes Ravi Vakil, Timothy Gowers, Geoffrey Irving, and Paul Christiano. MAISI is recruiting 10–30 faculty members for a January 2027 semester and 30–100 participants for a September 2027 Special Year, with members expected to relocate to the Bay Area for one or two semesters. The institute argues that AI safety needs more rigorous ways to define, measure, and reduce catastrophic risk, rather than relying only on subjective expert assessments. Its research will examine what premises could support high-confidence claims that an AI technology will not cause a catastrophe, while also considering whether some AI paradigms or safety-verification tasks could be provably impossible, chaotic, or computationally intractable. MAISI says mathematical work will complement, not replace, regulation, infrastructure hardening, and safety-oriented training. The institute will operate like a mathematical research institute, bringing mathematicians together with AI safety experts and sharing developing ideas through living documents before formal publication.