What it really takes to build safe, competent AI

A discussion with Ava Fitoussy and Liu Zhang, co-authors of “Safety Without Over-Refusal: Toward ‘Safe and Helpful’ Model Behavior Through Competence Under Constraint,” on designing AI systems that remain safe without defaulting to over-refusal. Moderated by Mark Esposito, the session focuses on competence under constraint and its role in building trustworthy, usable AI.