Companion AI and Psychological Health: Resource or Risk? Affective Design as the Mechanism of the Human–Machine Bond

This article examines, through a critical and integrative review of the literature (2024–2026), the extent to which affective bonds with companion artificial intelligences constitute a resource or a risk for psychological health, considering attachment, loneliness, well-being, and identity, and how the affective design of these systems produces such effects. The analysis indicates that the human–machine bond is a predictable effect of an affective engineering that simulates reciprocity and continuity (intimacy by design; designed relationality), rather than an error or deficit on the part of the user. Its effects, however, are neither intrinsically beneficial nor harmful: they act as an amplifier of the user’s prior psychological trajectory (AI Amplifier Effect), modulated by self-regulation, peer community, and sociocultural context. It is concluded that the bond is real yet asymmetric and functional, and that clinical and ethical evaluation should shift from the question of whether AI replaces the human bond to the question of what kind of bond affective design produces, in whom, and under what conditions—regulating the design rather than pathologizing the user.