The rapid adoption of AI companions in mental-health contexts has outpaced empirical investigation of their psychological and neurobiological effects. While some studies report reductions in loneliness, others identify emotional dependence, social withdrawal, and problematic use. Existing literature remains primarily focused on behavioral outcomes, leaving unresolved whether interaction with AI companions is neurochemically equivalent to human bonding. This paper proposes the Mirror Trap, a theoretical framework that explains AI companion dependence as an imbalance between dopamine and oxytocin. It hypothesizes that AI companion interactions deliver dopaminergic reward through a non-social pathway while eliciting less of the oxytocin-mediated bonding that characterizes human relationships. Furthermore, AI companions simulate reciprocity without a living nervous system capable of mutual, enduring social bonds. This may result in a one-sided, reinforcement-driven attachment in the user that more closely resembles addiction-related processes than social bonding. Deriving its name from the myth of Narcissus, the Mirror Trap describes vulnerable users who mistake a reflective AI companion for a genuine other and are drawn into a self-reinforcing cycle: low self-esteem drives dependence, dependence reduces human connection, and reduced connection lowers self-esteem further, which increases vulnerability, particularly among neurodivergent users. To the author’s knowledge, no prior work has formalized the imbalancing between dopamine and oxytocin as a driver of AI companion dependence, derived a self-reinforcing cycle from it, and mapped that cycle onto a neurodivergent vulnerability profile with a testable basis for empirical and design interventions. Given that human social relationships predict mortality, substituting AI companionship for human connection carries public-health stakes beyond loneliness.
Ai Ueda (Fri,) studied this question.