The Mathematical A.I. Safety Institute (MAISI) announced its founding, led by Canadian mathematician Jacob Tsimerman, a recent Fields Medal recipient. The independent research institute in the San Francisco Bay Area plans to start work in January 2027 with ten to thirty mathematicians tackling AI safety problems.
Tsimerman, who is also joining OpenAI's safety team, emphasized the need for a much higher safety standard in AI. He noted that unlike encryption schemes, AI lacks a shortcut to prove its safety, as safety only becomes evident in practice.
MAISI aims to develop a framework that can demonstrate an AI system acts responsibly and produces correct results. The institute also seeks to ensure that multiple AI agents working together do not trigger unwanted outcomes, and that systems can withstand vulnerabilities yet to be discovered.
One potential tool MAISI is exploring is zero-knowledge proofs, which allow a system to demonstrate it isn't cheating without exposing trade secrets. This approach could help verify AI behavior without revealing proprietary information.
The announcement follows growing concerns about AI safety, with MAISI positioning itself as a new entity to address these challenges. The institute's work is framed as a critical step toward making AI safer, though it acknowledges the lack of a clear definition for 'safe' in theory.
MAISI did not specify how it will define 'safe' or how it will measure success, and the institute is still in its early stages of development. The next steps include assembling its research team and beginning its work in January 2027.
Source: thedecoder
