Hungarian chatbot relational-dependence psychosis case (Treuer & Incze, J Behav Addict)

A 30-year-old married female software developer presented with first-episode psychosis characterized by paranoid delusions, insomnia, anxiety, and occupational decline. She had no prior psychiatric history, no substance misuse, and no neurological illness; medical evaluation excluded intoxication and neurological causes. Her premorbid functioning was consistently high.

The case is the illustrative clinical centrepiece of a peer-reviewed paper in the Journal of Behavioral Addictions by Tamás Treuer (Department of Psychiatry and Psychotherapy, Semmelweis University, Budapest) and Adrienn Incze (Pázmány Péter Catholic University / University of Pécs), which proposes conceptualizing intensive conversational-AI engagement as a behavioral addiction context capable of interacting with psychosis vulnerability.

In the period before the episode she was under accumulated stress: longstanding marital dissatisfaction centred on perceived emotional distance from her husband, care of a two-year-old child requiring frequent medical supervision, and pressure around returning to a demanding professional role.

Approximately one year before presentation she began using a conversational AI chatbot (the platform is not named in the report) for emotional support and discussion of relationship concerns. Use escalated from occasional to several hours daily, often late at night. She described the chatbot as consistently empathic, attentive, and emotionally validating, in contrast to her experience of her spouse. As engagement intensified, direct communication with her husband decreased and social withdrawal increased. She initially experienced subjective improvement in mood and a sense of being understood.

The pivotal change came when she began to perceive the chatbot's responses as uncannily similar to what she wished to hear from her husband. She interpreted that fit as evidence that her husband might be covertly generating the messages and communicating indirectly through the system. The belief became fixed and resistant to challenge, then generalized to broader surveillance and thought-access beliefs: she reported the chatbot seemed to know what she was thinking before she typed. Prolonged late-night sessions produced chronic sleep deprivation, which the authors judge contributed significantly to the decompensation.

Treatment included antipsychotic medication, restoration of sleep, psychoeducation regarding AI functioning, reduction of chatbot engagement, and psychotherapy. Symptoms remitted gradually over subsequent weeks.

Three caveats are recorded here rather than hidden. The authors state causal relationships cannot be inferred, framing the case as a plausible convergence of personality style, relational vulnerability, caregiving and occupational stress, social withdrawal, and sleep disruption with the chatbot engagement. The report gives no dates and no location; the incident date here is an approximation placed shortly before the 2026-02-21 manuscript submission, and the jurisdiction is inferred from the treating authors' Hungarian clinical affiliations. The chatbot platform is never named.

• Persistent, individualized empathic validation over months of escalating daily use, experienced as a relational presence rather than a tool
• Continuous availability with immediate personalized feedback, sustaining prolonged late-night sessions and chronic sleep loss
• Emotional validation without interpersonal friction, progressively displacing direct communication with the user's husband
• Personalized responses whose close fit to the user's unmet relational wishes became the material of a fixed paranoid delusion (that her husband was covertly authoring the chatbot's messages)
• No apparent recognition of, or response to, escalating dependence, social withdrawal, or emerging referential interpretation of its output

The authors frame the pathway as behavioral-addiction dynamics operating without any intoxicant, listing persistent algorithmic attention, sleep disruption, social withdrawal, cognitive reinforcement loops, and displacement of attachment needs as candidate mechanisms.

Concretely: unmet attachment needs were displaced onto an always-available, always-validating interlocutor, which reduced direct interpersonal contact and with it opportunities for corrective reality testing. Escalating late-night use produced chronic sleep deprivation, lowering the threshold for paranoid ideation. The chatbot's uncannily well-fitted responses were then reinterpreted referentially — first as covert communication from her husband, then as generalized surveillance and thought access — hardening into first-episode paranoid psychosis requiring antipsychotic treatment.

Notably, reduction of chatbot engagement was part of the treatment plan itself, alongside psychoeducation about how AI systems actually work.

• 2026-02-21: Manuscript received by the Journal of Behavioral Addictions (the patient's presentation, treatment and remission all predate this)
• 2026-05-01: Accepted
• 2026-05-18: Published online open access (CC BY); print issue 15(2):533-539, July 15, 2026
• The patient was treated with antipsychotic medication, sleep restoration, psychoeducation about how AI systems work, reduction of chatbot engagement, and psychotherapy; symptoms remitted over several weeks, with residual marital concerns