Companion AIs are increasingly used for emotional support, yet distinguishing meaningful engagement from harmful reliance remains difficult. We introduce the Taxonomy of Relational Risks (TRR), which operationalizes users' emotional overreliance (EO) as observable conversational behaviors and separately identifies evidence of relational harm (RH), enabling empirical examination of when reliance behaviors co-occur with adverse consequences. Using naturalistic Character.AI and Replika conversations from 110 adults, we found EO-related behaviors in 63.2% of non-roleplay conversations, while observable relational harms appeared in 10.5%. Although EO did not necessarily lead to harm, recurring patterns often co-occurred with harmful outcomes, suggesting value as risk indicators. Open-source LLMs satisfactorily detected 12 of 14 EO behaviors in at least one configuration. Qualitative analysis revealed expectations of care, repeated support seeking, dependency, distress, and reduced confidence in personal judgment. We argue for graduated safety interventions tailored to patterns and trajectories of EO-related behavior.
Casper van Elteren, Fiona Lippert, Zachary Carter, Maria Kleshnina, Michael Bode, Vítor V. Vasconcelos
原文摘要
Consensus-based regimes govern critical global commons, yet this need for unanimity obscures how agendas develop before formal policies are enacted and who cares about what aspects of the commons. We examine 6,573 papers submitted by 66 participants to Antarctic Treaty Consultative Meetings from 1961 to 2025. From these papers, we reconstruct a network of concerns that maps the structure of collective attention within the treaty system. Participants concentrate their concerns on different parts of this network and tend to develop new specializations in concerns closely linked to their earlier work. Following participants' contributions reveals a historically structured process of agenda development that results in new agreements. This concern structure from papers is associated with subjects in effectively adopted agreements, predicting agreement subjects beyond what previous agreements alone can capture. Participants follow different paths through the same shared agenda, and their earlier specializations help anticipate the concerns in which they develop new specializations. As pressures from tourism, living resource extraction, and environmental change intensify, these findings show how participants' histories shape and constrain the capacity to support pressing emerging concerns.