MAISI

Overview

The Mathematical AI Safety Institute (MAISI) is a Bay Area–based research institute announced in September 2026 with operations starting January 2027. Its declared aim is to pursue formal safety proofs for AI systems — an approach that pattern-matches to cryptographic proof of security rather than empirical evaluation. Scientific Director Jacob Tsimerman is the 2026 Fields Medallist; Andrew Critch is Executive Director. The inaugural cohort is 10–30 mathematicians, with a “special year” in September 2027 expanding to 30–100.

Timeline

• 2026-09-12-AI-Digest — MAISI announced (Sep 8) with January 2027 operations; The Decoder frames the ambition as “prove-AI-is-safe-the-way-cryptographers-prove-codes-are-unbreakable”. Small formal-methods bet added to the safety portfolio — complementary to existing empirical labs (METR, ARIA, Apollo), not a paradigm shift.

Key Developments

1. Institutional launch: Announced 2026-09-08 with a Fields Medallist at the helm — the strongest single seriousness signal for a formal-methods AI-safety programme to date. Tsimerman also joins OpenAI‘s safety team; the roles are complementary.
2. Scale calibration: 10–30 mathematicians for the inaugural cohort is roughly an order of magnitude smaller than existing empirical AI-safety labs — this is a bet on a hard problem, not a displacement signal for empirical eval.