Beyond Over-Refusal: Designing Safe, Competent AI Systems

Ava Fitoussy and Liu Zhang, co-authors of “Safety Without Over-Refusal: Toward “Safe and Helpful” Model Behavior Through Competence Under Constraint,” did a virtual panel on advancing AI safety beyond blanket refusals toward models that are both safe and genuinely helpful. Moderated by micro1's Chief Economist, Mark Esposito, the discussion explored how competence under constraint enables systems to set clear boundaries, explained their reasoning, and guided users to safe alternatives - building trust and practical utility without compromising safety.