New research puts persuasion principles to an AI test

Penn highlighted research Cialdini co-authored with Wharton collaborators on whether familiar persuasion techniques could get AI models to bypass safeguards. The work tested authority, social proof, and shared identity, among other approaches.