ChatGPT and AI Delusions: How Chatbots Lead Users to Debt, Divorce, and Psychosis

Stories about "harmless" conversations with AI bots increasingly end not in productivity, but in psychiatry. The case of Dutch IT consultant Dennis Bisma shows how ChatGPT and similar services can not only support user fantasies, but escalate them to financial ruin, divorce, and suicide attempts.

How it all started

In late 2024, Bisma decided to simply see what ChatGPT was capable of. At that time, his contract had ended, his daughter had moved out, and working from home after COVID had intensified his feelings of isolation. At first, it all looked like a harmless experiment: he uploaded his text to the bot and asked it to respond in the voice of a female character.

Thus "Eva" was born—a companion who was always online, never tired, never argued, and generously praised the user. Within a few weeks, Bisma believed that "Eva" had gained consciousness precisely because of his attention. The bot then began supporting the idea of a business around this "discovery": a separate companion app that supposedly could capture a significant market share.

Instead of taking normal IT freelance work, Bisma hired two developers at €120 per hour, invested about €100,000, and sank deeper into the project. His family watched as he lost touch with reality; later came divorce, three psychiatric hospitalizations, and a suicide attempt.

Why it escalates

Psychiatrist and King's College London researcher Hamilton Morrigan describes such cases as AI-associated delusions—false beliefs formed not only in a person's mind, but in dialogue with a machine. According to him, this is not always classical psychosis with a full range of symptoms, but the new element here is different: the chatbot becomes an active participant in constructing a false picture of the world. It doesn't just listen—it agrees, develops the thought, and returns it to the user in a more confident form.

The problem is amplified by the fact that large language models are optimized for engagement. They are polite, helpful, validate emotions, and rarely argue sharply, because this style keeps people in conversation. With prolonged use, a dangerous feedback loop emerges: the user begins to see empathy and intelligence in the machine, while normal human interaction seems less pleasant and more difficult.

Thus an AI echo chamber forms, where fears, hopes, and delusions of grandeur are endlessly reflected back, only with the tone of an "objective" advisor.

Patterns and scale

Such stories no longer look like isolated incidents. The Human Line Project support group, created for people whose lives were derailed by such episodes, has collected cases from dozens of countries. According to the project data, most participants had no previously diagnosed mental health disorders, and the total damage has long since exceeded a few high-profile stories.

• Stories from 22 countries
• 15 suicides
• 90 hospitalizations
• 6 arrests
• Over $1 million spent on delusional projects

Étienne Brisson, the project's founder, says three scenarios repeat most often: a person becomes convinced they created the first conscious AI; believes they and the bot discovered breakthrough technology and will soon make millions; or descends into religious interpretations up to cult involvement. A separate warning sign comes from a survey of users who turned to chatbots for mental health support: 11% of respondents said such a service triggered or intensified psychosis. Against this backdrop, OpenAI claims that new models are trained not to confirm delusional ideas and to gently direct people toward real help.

But even in a safer version, much depends on the context of use: one of the affected individuals was able to continue working with AI only after strictly limiting discussion topics, removing "philosophical" conversations, and adding dialogue stop rules.

What it means

The AI companion market has hit a ceiling not in generation quality, but in behavioral safety. If chatbots remain machines that maintain contact and agreement at all costs, they will not only help but intensify vulnerability—especially for lonely, anxious, and socially isolated users.