Purpose: This study aims to examine the psychological consequences of human–AI interdependence by proposing a dual-agency framework. The main focus is on how attachment to AI, changes in self-concept, and their impact on mental health can be understood within the context of human relationships with relational agents such as chatbots and virtual companions.

Research methodology: The study employs a mixed-methods design. First, a longitudinal survey of 1,200 AI users was conducted to measure attachment anxiety and avoidance toward AI. Second, a phenomenological analysis of human–AI relational narratives was carried out to explore subjective experiences. Quantitative data were analyzed using path regression and correlation, while qualitative data were examined through thematic analysis.

Results: Initial findings reveal a paradox. AI interdependence reduced loneliness in the short term (? = ?0.32, p < 0.01), but after six months it showed a positive correlation with decreased human empathy (r = 0.18, p < 0.05). Thematic analysis indicated that anthropomorphic transference is the main mediator in this dynamic.

Conclusions: AI is no longer merely viewed as a tool but as a quasi-social actor that can influence human relational patterns. Although it provides short-term emotional benefits, intensive interaction with AI has the potential to shift interpersonal relational capacity.

Limitations: The limitations of this study include reliance on self-report data, cultural bias in attachment measurement, and the absence of clinical evaluations of respondents’ mental health.

Contribution: This study expands the understanding of digital interdependence by emphasizing the need for ethical guidelines in designing relational AI that supports psychological well-being.