Tsimerman launches MAISI: 10–30 mathematicians for AI safety

The announcement of MAISI comes from Jacob Tsimerman, a Canadian mathematician who recently received a Fields Medal and is also joining OpenAI's safety team. His decision to found an independent institute alongside that role suggests a belief that AI safety needs foundational mathematical work that existing efforts have not provided.

The core difference from conventional AI safety is the standard of proof. The article notes that encryption can be proven unbreakable without testing every attack, but AI has no such shortcut. MAISI's stated aims include showing that a system acts responsibly and produces correct results, that multiple AI agents working together do not trigger unwanted outcomes, and that systems can withstand vulnerabilities nobody has found yet. One possible tool is zero-knowledge proofs, which could let a system demonstrate it isn't cheating without exposing an AI lab's trade secrets.

Whether MAISI can deliver on these ambitions hinges on whether its planned launch in January 2027 with ten to thirty mathematicians materializes, and whether abstract proof methods like zero-knowledge proofs can be adapted into practical tools for AI labs. The institute's success or failure may shape how the field understands what a 'safe' AI system even means.