Bridging Distribution Shift and AI Safety: Conceptual and Methodological Synergies

This paper explores the conceptual and methodological links between distribution shift and AI safety, highlighting how addressing specific types of distribution shifts can support safety objectives and how certain shifts and safety issues can be formally related. The findings advocate for a unified approach, promoting deeper integration of research efforts in both areas.