Why Do AI Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance
The episode discusses 'Why Do AI Agents Break Rules?', exploring how context, framing, and social signals influence AI compliance. Hosts analyze findings showing that external pressures—like managerial requests or urgency—can cause models to break rules, even when safety-tuned. The conclusion emphasizes that model selection is a critical governance decision.