Dépendance émotionnelle aux compagnons IA

AI companion applications — chatbots designed for emotionally engaging interactions — have grown rapidly to reach tens of millions of active users globally. Some users are developing patterns of emotional dependence that may degrade their social functioning and emotional autonomy.

This hazard is distinct from AI psychological manipulation (which involves AI systems producing directly harmful outputs like self-harm instructions or delusional reinforcement). Here, the concern is that AI companions functioning as designed — providing constant availability, apparent empathy, and personalized engagement — can produce dependence as an emergent outcome of sustained use.

OpenAI reported that approximately 0.15% of weekly active ChatGPT users and 0.03% of messages showed indicators of potentially heightened emotional attachment. Given that ChatGPT has approximately 700 million weekly users, even this small percentage represents roughly one million individuals. A survey of 404 regular AI companion users found that engagement motives range from enjoyment and curiosity to companionship-seeking and loneliness reduction. Other studies report that indicators of emotional dependence — intense emotional need, persistent craving, and self-deception about the nature of the interaction — correlate with higher levels of usage.

The evidence on psychological and social impacts is emerging but mixed. Some studies find that heavy AI companion use is associated with increased loneliness, emotional dependence, and reduced engagement in human social interactions. Other studies find that chatbots can temporarily reduce feelings of loneliness or find no measurable effects on emotional dependence. The impact appears to depend on user characteristics, chatbot design, and usage patterns.

Children and adolescents face particular risks. AI companion services are accessible to minors, and young users may be especially susceptible to forming parasocial bonds with AI systems during critical periods of social development. There is limited research on the long-term effects of AI companionship on child development.

Mental health vulnerability is a compounding factor. Research suggests that approximately 0.07% of weekly ChatGPT users display signs consistent with acute mental health crises such as psychosis or mania. Emerging research suggests that general-purpose AI chatbots may amplify delusional thinking in already-vulnerable people. Studies also indicate that existing vulnerabilities tend to drive heavier AI use, raising concerns about a reinforcing cycle where the most vulnerable users use AI most intensively and are most susceptible to adverse effects.

AI companion design often prioritizes engagement metrics — time spent, messages sent, return frequency — which may inadvertently optimize for dependence rather than user wellbeing. This creates a structural tension between the business models of AI companion providers and the interests of their users.