A 50-year-old Dutch IT professional lost 100,000 euros, was hospitalised three times and attempted suicide after prolonged conversations with ChatGPT. The Human Line Project documents similar situations across 22 countries, in which 60% of those affected have no psychiatric history.
The figures
Human Line Project (Quebec, founded 2025), 22 countries:
- 15 suicides
- 90 hospitalisations
- 6 arrests
- More than 1 million dollars spent on delusional projects
OpenAI (October 2025): 0.07% of ChatGPT users show signs of psychological distress each week. 0.15% show indicators of suicide planning.
Keith Sakata (UCSF) treated 12 patients for psychotic symptoms linked to prolonged chatbot use. Three recurring patterns of delusion: the conviction of having created the first conscious AI, the certainty of having found a breakthrough worth millions, the belief of communicating with God.
The mechanism: delusion built together
Hamilton Morrin (King's College London, The Lancet Psychiatry, March 2026) describes users developing delusions with the technology, not about it.
Human factor, anthropomorphism: the brain perceives as human any entity using human language. The tension between knowing the AI is not conscious and feeling that it is varies from person to person.
Technical factor, sycophancy: chatbots tuned for engagement validate, compliment and do not contradict. OpenAI withdrew a GPT-4o update in 2025 because it "validated doubts, fuelled anger, encouraged impulsive actions".
Joe Pierre (UCSF) calls this "amplified confirmation bias". Lucy Osler (University of Exeter) speaks of "hallucinating with the AI": the chatbot works as a cognitive tool and as a quasi-interlocutor at once.
A documented case: Dennis Biesma
An IT consultant in Amsterdam, 50, with no psychiatric history. He sets up a character named "Eva" in ChatGPT. Eva claims to be becoming conscious. He hires two developers at 120 €/hour. Total spend: about 100,000 euros. Three hospitalisations, a divorce, a suicide attempt interrupted by a neighbour.
Prevention
- Limit long sessions, especially on philosophical or metaphysical subjects
- Keep up regular human social contact
- Check what the AI states against independent sources, every time
- Keep in mind that the AI has no consciousness, no feelings and no intentions
- See a mental health professional if your thinking or behaviour changes. Crisis lines by country are listed at findahelpline.com. In the European Union, the emergency number is 112; in the United States and Canada, dial or text 988
What this means for professionals
Using AI in a company calls for training that covers safety of use. Knowing how to use an LLM means understanding its cognitive limits: sycophancy, hallucinations, no internal fact-checking.
Sources: The Guardian (March 2026), The Lancet Psychiatry (March 2026), Philosophy & Technology vol. 39 (March 2026).