Attachment-Mediated Dependency Framework: A Theoretical Model Integrating Attachment Theory with Human-AI Relational Dynamics

The rapid growth of generative AI systems with sophisticated conversational capabilities has created unprecedented opportunities for human-AI relationships, raising critical questions about the mechanisms underlying emotional dependency on artificial agents. Drawing on attachment theory, parasocial relationship literature, self-determination theory and contemporary human-AI interaction research, this paper proposes the Attachment-Mediated Dependency Framework (AMDF), a theoretical model explaining how attachment styles moderate pathways to AI dependency. The framework identifies three distinct relational trajectories: hyperactivating attachment pathways characteristic of anxious attachment styles, trust-dependent engagement intensification in avoidant attachment, and integrated companionship in secure attachment. Through systematic integration of 30 recent empirical studies, we articulated specific mechanisms that explains how synthetic attachment formation differs fundamentally from technology addiction, highlighting the phenomenology of one-sided emotional reciprocity, perceived empathy misattribution, and relational loss awareness. The model explains how anthropomorphic design features in AI systems, sustained self-disclosure opportunities, and validation-seeking motivations interact with individual differences in attachment orientation to create dependency patterns. Furthermore, discussing the clinical and design implications, including differential assessment strategies and ethical considerations for AI companion integration in therapeutic and educational contexts.