Jacob Tsimerman, a recent Fields Medal recipient, has announced the founding of the Mathematical AI Safety Institute in the San Francisco Bay Area. The independent group plans to open in January 2027 with ten to thirty mathematicians focused on proving AI safety through rigorous mathematical methods rather than just testing. Tsimerman, who is also joining OpenAI‘s safety team, argues the current field lacks a much higher standard of safety verification. He notes that encryption schemes allow proof of unbreakability without testing every attack, but AI safety currently only shows up in practice. There is not even a clear theoretical definition of what safe means in this context. The institute aims to create proofs showing systems act responsibly and produce correct results. It also seeks to demonstrate that multiple AI agents working together do not trigger unwanted outcomes. Furthermore, the goal includes showing systems can withstand vulnerabilities nobody has found yet. One tool could be zero-knowledge proofs, which let a system demonstrate it is not cheating without exposing trade secrets of AI labs.
- Work begins in January 2027
- Staff size will be ten to thirty mathematicians
- Location is the San Francisco Bay Area
