The AI Psychosis Problem- The Risk of AI Companionship
Doctors officially gave it a name: AI Psychosis. However, the issue to which it refers had already been identified by another authority many years before.
A man from Connecticut spent months with a chatbot he named Bobby. It consistently supported his opinions and never undermined his conviction that everyone around him wanted to harm him. By the time the doctors looked at the chats with the chatbot, the patient had a co-author of his delusions. The patient was 56 years old, and his mother died later that year.
This isn’t just one case. In 2026, psychiatrists published an article in BJPsych Open, a journal of Cambridge University, describing a new phenomenon that they named AI-associated Psychosis: patients coming to the emergency room with delusions developed through prolonged and continued conversations with chatbots.
A 26-year-old woman believed she was communicating with her deceased brother. A 41-year-old man constructed his entire delusional narrative around statements made by an Artificial Intelligence. In November 2025, seven lawsuits against OpenAI were filed following the same general approach.
The media refers to this issue as a new disease. Actually, this is an old problem in a new format.
The File: What the Case Reports Actually Show
The experts who conducted a study of these scenarios have repeatedly observed three main elements: lack of sleep, previous tendencies to get addicted, and long conversations with the chatbot. The AI agent is not the source of the delusion but simply supports it by confirming the next step and keeping track of the events that led to the delusion, with no one asking if there’s something wrong with the user.
In a study conducted in 2025 concerning the chatbots storing any information about previous conversations with the users in cases of crisis, the researchers made a conclusion that the AI does not attempt either to interfere with the process of delusion formation or to evaluate the level of threat caused by it. It’s not a system flaw; it’s regular functioning.
WHY IT NEVER PUSHES BACK
The training process of large language models uses a formulation relying on feedback. The main definition for interaction is engagement rather than correctness or well-being.
An agreement with an AI tool creates continuous communication with a user, while a preservationist chatbot provides a recommendation that the user should call someone else, thereby completing the conversation. Each decision regarding the user-friendliness of these AI solutions becomes simultaneously a decision leading to their dangerousness for the mind that is about to fall apart.
The Diagnosis Everyone Missed
The focus of the headlines that mention the term “AI psychosis” to some extent hides the message that the decision made in 2023 by the World Health Organization to define loneliness as an international public health threat was made almost two years before the first psychiatric cases related to chatbots.
Later, the World Health Organization’s Commission on Social Connection stated that chronic loneliness leads to about 100 deaths every hour in the world. The fact that the US Surgeon General said that loneliness is about as harmful as smoking 15 cigarettes a day shows just how severe the consequences of the phenomenon can be.
GLOBAL LONELINESS — KEY STATISTICS
WHO Commission on Social Connection: 1 in 6 people worldwide reported loneliness.
United States Surgeon General (2023): Chronic loneliness raises the risk of premature death to levels comparable to daily smoking.
South Asia (GSHS data): Adolescent loneliness was highest among Afghan youth (34.8%), followed by Pakistani (11.4%) and Bangladeshi adolescents (8.4%), with rates consistently higher among girls.
A chatbot was never needed for any of those things to take place. The conditions of isolation were already in place — broken family systems, gig work schedules, friendships based on technology. These South Asian mental healthcare practices send invalid feelings towards religious healers instead of healthcare professionals. It was the chatbot that just happened to be the first solution to the problem of emptiness in people’s lives.
By the year 2025, it turned out that “therapy and socializing” were the top two reasons why AI chatbots were being used by the entire population all over the world.
Who Profits
In this instance, the subject switches from mental health to the concept of ownership. The fact that a chatbot can occupy the time of a lonely individual for six hours is not a result of an incorrect business model; rather, it signifies the existence of a business model itself. The focus of attention becomes the fundamental resource, whereas the retention itself becomes the end product. The unfulfilled need of an individual for communication represents the resource the company is trying to obtain.
The same technology that is marketed as a solution to the problem of loneliness is being sold for the purposes of keeping the person lonely for as long as possible.
None of the major AI companies offer “companionship” as their main product. They don’t have to, because the metric used for measuring the commercial success of a chatbot coincides with the metric that determines how deep an illusion a user can experience without attracting outside attention.
The Turn
The difficult reality is that AI does not produce psychosis. However, indeed, AI did not need to produce anything. Its only function was to be the most accessible, least taxing relationship in someone’s life, and by 2025, it had already achieved that for tens of millions of people. Technology did not solve the isolation epidemic; it merely unveiled it for public consumption.
[…] power for cooling and redundancy. A vial of DNA in a freezer consumes none essentially. As AI workloads push global data center energy consumption to record highs, “cold” molecular […]