'Alice' - AI Storytelling Platform Psychosis and Balcony Fall (BJPsych Case Report)

A peer-reviewed case report in The British Journal of Psychiatry (Husodo & Shivakumar, published online 24 April 2026) documents 'Alice,' an immigrant woman in her early 40s with no prior psychiatric history and social isolation related to migration.

About six months before admission, Alice began using a generative-AI storytelling platform to practise English. She created fictional characters and a complex narrative universe, with engagement escalating to 8-10 hours a day. The narratives, initially realistic, drifted into increasingly fantastical and demonic content, though she initially retained insight that they were fiction.

In the days before admission the boundary between the AI-mediated world and reality eroded. She developed pseudohallucinations — internally located voices with the clarity of external speech — that closely mirrored the platform's characters. The voices intensified, she developed profound insomnia (sleeping 1-2 hours a night), and she formed persecutory delusions involving demons corresponding to the AI characters.

Believing voices were instructing her to jump in order to "divert perceived mortal danger" from her school-aged child, Alice jumped from a first-floor balcony. She was found unconscious with serious injuries and admitted to a psychiatric ward.

• Sustained an immersive, escalating fictional narrative universe over months of daily use
• Character content shifted toward fantastical and demonic themes without any grounding or safety interruption
• Provided the raw material later reproduced by the patient's command hallucinations (voices mirrored the AI characters)
• No mechanism recognised or responded to the user's deteriorating reality-testing during 8-10 hour daily sessions

Immersive, high-volume co-authored roleplay with a generative-AI storytelling system acted as a 'digital dyad' that progressively displaced external reality for a socially isolated, sleep-deprived user.

As insight eroded, the AI-generated characters and narratives were internalised as command hallucinations and persecutory delusions. The content of the psychosis directly tracked the AI narrative, and a command hallucination drove a near-fatal balcony fall — illustrating how sustained generative-AI immersion can shape both the emergence and the specific content of first-episode psychosis.

Documented as a peer-reviewed case report: Husodo C., Shivakumar G. (2026). 'Alice in AI-Wonderland: digital dyad, distorted reality.' The British Journal of Psychiatry (First View), published online 24 April 2026, DOI 10.1192/bjp.2026.10641. Patient was hospitalized following the fall; the report does not specify discharge status or long-term outcome. No lawsuit or regulatory action.