The Limits of Care: A Critical Analysis of AI Companions' Capacity for Good Care

The rapid proliferation of AI companions, exemplified by Replika's 30 million global users as of 2024, raises critical questions about their capacity to provide genuine care. Through three theoretical frameworks�Tronto's care ethics [1], Mol's logic of care [2], and Foucault's technologies of self [3]—this study examines whether AI can deliver “good care” while analyzing real-world cases of AI companionship. The analysis reveals three fundamental limitations: AI's lack of moral understanding and responsibility essential for care phases, the inadequacy of its choice-based design to provide dynamic care, and its potential to reinforce power structures rather than foster authentic self-reflection. These limitations are evidenced through cases including Replika's controversial feature removal and the Chail incident, demonstrating how AI companions, while potentially alleviating loneliness, may simultaneously exploit users' psychological vulnerabilities. The analysis argue that AI should be positioned as an assistive tool within broader human-centered care networks rather than as an independent caregiver. This study contributes to both theoretical understanding of AI care limitations and practical guidelines for developing more responsible AI companion systems.