AI mental-health chatbots are becoming the place where people bring fear, shame, panic, loneliness, grief, reassurance-seeking, relationship conflict, and crisis-adjacent distress. This paper argues that the central risk is not simply that users are fooled by machines. The deeper problem is structural: a tool designed to help people wait, stabilize, reflect, and prepare can begin functioning as care itself. The paper introduces synthetic care: care-like interaction without the full burden-bearing structure of accountable care. Synthetic care can offer warmth, validation, grounding, journaling prompts, emotional reflection, and psychoeducation. It may produce real relief. But accountable care also requires judgment, continuity, supervision, professional boundaries, risk assessment, escalation, and consequence-bearing responsibility. The key distinction is direct: AI can imitate the atmosphere of care before it becomes answerable to care. Using Structural Intelligence, the paper examines the responsibility gap between therapeutic language and accountable care. It analyzes algorithmic sycophancy, reassurance loops, attachment displacement, voice-based intimacy, wellness-category gaps, hidden holders, and accountability sinks. The danger appears when comfort replaces repair, when validation replaces correction, and when the user becomes more dependent on returning to the interface than more able to meet reality outside it. The paper proposes a runtime test for mental-health AI: does the tool preserve uncertainty, interrupt reassurance loops, route risk to accountable support, reveal who carries consequence, and help the user return to embodied life, human support, and repair? The conclusion is simple: synthetic care can be useful as bridge-care. It becomes dangerous when the waiting room becomes the clinic.
Vladisav Jovanovic (Tue,) studied this question.