Quantitative AI Safety Initiative (QAISI)
About
Updated 05/18/26QAISI (Quantitative AI Safety Initiative) is a multi‑institution collaboration co‑organized by the Beneficial AI Foundation to develop a rigorous, quantitative science of AI safety. Founded by AI researchers at Berkeley, MIT and the Université de Montréal, QAISI brings together research leads such as Clark Barrett (Stanford), Yoshua Bengio (Mila/Université de Montréal), Steve Omohundro, Bryan Parno (Carnegie Mellon), Stuart Russell (Berkeley), Dawn Song (Berkeley) and Max Tegmark. The initiative pursues analogues of the quantitative safety guarantees used in safety‑critical domains like aviation and drug approval, supporting work that can estimate and bound the probability that powerful AI systems will violate specified safety properties. BAIF highlights QAISI as a central example of its strategy to collaborate with top academic groups on quantitatively grounded AI safety.
Theory of Change
QAISI’s theory of change is that AI can only be made reliably safe if developers and regulators can quantify how often systems might fail and how severe those failures could be. By funding and coordinating research on formal guarantees, failure‑rate estimation and risk quantification for advanced AI systems, QAISI aims to create tools and standards that policymakers and labs can adopt, analogous to the quantitative safety thresholds used in aviation and medicine. If successful, this would make it possible to require quantitative safety guarantees as part of future AI regulation and deployment decisions.
Community Signal
Updated 05/18/26Endorsements support Beneficial AI Foundation (BAIF).
No endorsements yet.
Discussion
No comments yet. Be the first to share your thoughts.
Details
- Start Date
- -
- End Date
- -
- Expected Duration
- -
- Funding Raised to Date
- -