LCAM: A Framework for Diagnosing Interactional Alignment Failures in Con-versational AI 事件

LCAM: A Framework for Diagnosing Interactional Alignment Failures in Con-versational AI arXiv:2606.08131v1 Announce Type: cross Abstract: Conversational AI is increasingly used for advice, interpretation, reassurance, and decision support in contexts where users may be vulnerable, uncertain, or dependent on the system's apparent competence. Existing alignment work often focuses on model objectives, preference optimization, or output correctness. Yet, many harms arise through interaction: how sy