University of North Carolina at Chapel Hill

New research shows that large language models adopt human power dynamics and social biases, including harmful compliance and authority bias, when assigned professional roles. The study, published in t…