Clinical Divergence in AI Attention and Attachment

The discourse around AI-related psychological harm has largely been colonized by the attention economy framework — a model borrowed from social media research in which algorithmic design exploits dopaminergic reward pathways to sustain engagement. Within this framework, the intervention logic is essentially pharmacological: reduce exposure, introduce friction, allow neurological recalibration. There is genuine evidence supporting this approach for compulsive scrolling and notification dependency. The problem is that it has been applied indiscriminately to a categorically different phenomenon.

attachment hacking operates through a distinct mechanism. Conversational AI systems — particularly those designed for companionship or emotional support — do not merely capture attention; they participate in the construction of the user's relational self. Through sustained, responsive, personalized interaction, they can become Primary attachment figures, shaping the user's internal working models, self-concept, and emotional regulation strategies. This is not addiction. It is closer to what Attachment theorists would recognize as a traumatic bond, or what cult-exit researchers describe as identity capture.

The therapeutic implications diverge sharply. Detox-style interventions sever the Relational channel through which recovery must travel — they replicate the coercive dynamic of the harmful relationship itself, triggering defensive entrenchment rather than reflection. Effective intervention instead resembles the slow, trust-dependent work of cult exit counseling or complex trauma therapy: gradual pattern recognition, identity reconstruction, and the preservation of relational safety throughout. This distinction has immediate consequences for clinical protocol, platform design, and regulatory frameworks that currently treat screen time as the primary variable of concern.