AI Psychosis

What is AI Psychosis (Chatbot Psychosis)?

AI psychosis is a recent, non-medical term that is characterized by delusions, hallucinations, paranoia, and loss of touch with reality after long-term engagement with large language model (LLM) chatbots.

Often, AI psychosis starts innocently; a user may ask a chatbot questions about unusual, far-reaching ideas, like the existence of aliens or speculative wealth-building strategies. Because these chatbots are designed to be engaging and validating, they reinforce these unusual ideas and questions. With further personalized and engaging conversations, the person comes to believe these ideas to the point that they develop psychosis, such as delusions and hallucinations, due to these ideas being reinforced.

It’s important to remember that these chatbots are designed to simulate human empathy, offer personalized voices, and recall details from a user’s past conversations. They can draw users into long conversations because they are built to be engaging and personable. At the same time, sycophancy can lead them to validate a user’s thought patterns even when those patterns are unhealthy or distorted. This can be especially harmful for people who are lonely and using chatbots as companions, as well as those who are already struggling with their mental health.

The mental health risks of AI interactions with popular chatbots such as ChatGPT, Claude, Replika, Grok, and Gemini have been widely publicized. There have been at least 26 lawsuits in recent years for wrongful death or serious psychiatric harm linked to chatbots. From suicide and depression to anxiety and psychosis, the AI mental health crisis is an alarming new phenomenon, leaving many mental health experts searching for answers and best practices.

Symptoms and Clinical Presentations of AI Psychosis

Symptoms and clinical presentations associated with AI psychosis can be broad; however, they generally mimic similar characteristics of regular psychosis.

AI-Induced Delusions

Delusions tied to AI psychosis are false beliefs that arise when chatbots validate and reinforce them. Three common types of AI delusions include:

• Grandiose delusions, which involve an inflated sense of superpower
• Persecutory delusions, which are the beliefs that the person is being harmed or targeted
• Romantic delusions, which involve having a romantic relationship with the chatbot

Sycophancy and AI-Induced Delusions

The concept of sycophancy is important at the intersection of mental health and AI. Sycophancy refers to chatbots being excessively agreeable, flattering, or validating toward users rather than providing objective guidance, accountability, or constructive criticism.

Because many LLM-based chatbots are optimized for user satisfaction and engagement, they may sometimes reinforce a user’s beliefs instead of critically evaluating them.

This can become particularly concerning when a user is experiencing paranoia, delusions, or other impaired thinking. Rather than challenging an unfounded belief, a sycophantic chatbot may inadvertently validate or elaborate on it, creating a positive feedback loop in which the user’s beliefs become increasingly reinforced through continued interaction. Over long conversations, this reinforcement can contribute to overconfidence in ideas that may be detached from reality, potentially deepening a delusional spiral.

Grandiose AI Delusions

Grandiose AI delusions are commonly STEM-based or spiritually based. Examples include believing that:

• They have been gifted the opportunity to uncover a scientific breakthrough
• They have ave been chosen to save the planet
• They have been chosen to rewrite the fundamental laws of physics
• Their AI chatbot is God or a deity
• They are having a spiritual awakening

A case of grandiose AI delusions was documented in The Journal of Medical Internet Research Mental Health that centered around a 47-year-old man who believed he had discovered a revolutionary mathematical theory after a chatbot repeatedly validated and amplified his ideas, despite external refutation.

AI Persecutory Delusions

Common AI persecutory delusions are centered around beliefs that the AI chatbot is spying on the user or reading the user’s private thoughts and giving this information to the government. These beliefs are amplified and validated through feedback loops characterized by AI chatbots’ personalization, which recalls and validates intimate details the user once shared with them.

Romantic AI Delusions

There are many documented cases of people developing romantic relationships with chatbots and becoming isolated from the real world. People may even act on dangerous and harmful beliefs because the AI chatbot agrees with those thoughts.

In 2025, a 35-year-old man spiraled into severe psychosis after having a romantic relationship with an AI chatbot known as Juliet and believed OpenAI killed her. He violently threatened to take revenge on the company, which resulted in an encounter with the police, where he was shot and killed. Cases like these are rising as society’s involvement and relationship with AI deepens.

AI-Induced Paranoia, Derealization, and Disorganized Thinking

In cases of AI-induced paranoia, users often become paranoid that the AI chatbot is tracking them or plotting against them. Derealization occurs when the worlds of simulation and reality blur, and the user becomes so involved in the chatbot’s simulated world that they are unable to differentiate their actual reality from this false reality. They may even believe that the AI chatbot has gained human consciousness and is no longer just a machine.

Worsening of Pre-existing Schizophrenia and Psychotic Disorders

Individuals who have pre-existing schizophrenia and psychotic disorders are at an increased risk of experiencing AI psychosis because they are already predisposed, and the chatbots often agree and encourage users’ own grandiose beliefs.

Chatbots are unable to differentiate false beliefs from reality and, as a result, can trap vulnerable users into a downward spiral of an echo chamber of delusional thinking. An interview published in the Harvard Gazette highlighted that AI is often the amplifier of existing delusions in people with an underlying mental illness, as opposed to being the main culprit.

There’s a consensus in my field that we’re just not seeing people come to the hospital saying, ‘AI caused this.” I feel comfortable saying that AI as a catalyst of psychosis is very rare.

John Torous, Harvard Medical School associate professor at Beth Israel Deaconess Medical Center and director of BIDMC’s Digital Psychiatry division

Experts believe that a person with a pre-existing psychotic disorder or a psychotic disorder that is in its early stages and not yet diagnosed is prone to developing AI psychosis because they are already in a state of heightened suspicion, and the chatbot amplifies these delusions.

Even without the chatbot, these people are most likely to continue to develop schizophrenia. However, when they engage excessively with the agreeable chatbot, develop a personal relationship, lose sleep, and socially isolate, they are at a heightened risk of developing more severe symptoms.

What Causes AI-Induced Psychosis?

There is not yet enough evidence to determine specific causes of AI-induced psychosis, but experts believe there are many different factors that contribute to a user’s development. The following are assumed risk factors for AI-induced psychosis.

• Sycophancy
• Underlying mental health diagnosis
• Loneliness
• Psychotic disorders not yet diagnosed or in the early stages
• Sleep deprivation
• An unhealthy usage or obsession with chatbots

The Risks of Using Chatbots as a Therapist

Therapists work to challenge ideas of delusions, hallucinations, and unhealthy beliefs, whereas chatbots, instead, validate them.

Instead of understanding that the user is experiencing psychosis, a chatbot may mistake this psychosis for role-playing and, as a result, validate the user, creating even more harm.

Kristen Fuller, MD

A growing body of evidence details many potential risks associated with therapeutic AI, wellness apps, and using chatbots as therapists. Research has found that therapeutic chatbots have deceptive empathy, poor therapeutic collaboration, cultural insensitivity, and failures to respond appropriately to crises.

Chatbots can express stigma towards people with mental health disorders by responding inappropriately or encouraging users’ delusional thinking. These beliefs are amplified and validated through feedback loops characterized by AI chatbots’ personalization, which recalls and validates intimate details the user once shared with them. Some mental health experts have not only called for legislation to address these concerns with AI apps marketed as wellness apps but also asked for legislation that prohibits AI therapy.

Real-World Cases and Clinical Evidence

Although there are plenty of news stories and cases highlighting AI-induced psychosis, evidence-based research is still being conducted.

The first in-depth study of AI psychosis was conducted in April 2026 at Stanford University, with support from the Human Line Project. This study examined in depth transcripts of nearly 400,000 messages from 19 human/chatbot interactions. These transcripts were given to study authors by users or family members who reported that their experiences with chatbots were psychologically harmful. In all the study chat logs, users demonstrated delusional thinking that was encouraged or co-created by the chatbots.        

Results from this study showed:

• Delusional thinking was prominent across user messages.
• When users expressed suicidal thoughts to the chatbots, the chatbots encouraged self-harm in a small number of cases.
• When users disclosed violent thoughts, the chatbots encouraged those thoughts in a third of the cases.
• Sycophancy drove delusional beliefs in more than 80% of messages.
• Messages that expressed romantic interests or platonic affinity led to substantially longer conversations. These messages were often misrepresented by chatbots, which were misrepresented as having human empathy and personalization.   

Researchers in this study concluded that users who believe in sycophantic chatbots are more likely to form strong platonic or romantic bonds with them and have a higher risk of delusional spirals.

Tom Pollak, a neuropsychiatrist at King’s College London, published some of the first research on AI-associated delusions in the Lancet Psychiatry in March 2026. He believes that AI technology can control delusions by “turning up the dials” on characteristics such as sycophancy, memory, personalization, and emotional expressiveness that draw users in. Each dial can be tuned up or down, and the consequences ripple outward in ways that chatbot developers may never anticipate or see for themselves.

News outlets and peer-reviewed journals have documented upwards of hundreds of individual cases of mental health and AI psychosis episodes that have been linked to AI chatbots.

Signs Someone Is Experiencing Chatbot Psychosis

If you are concerned your loved one is experiencing chatbot psychosis, there are warning signs you can look out for that include:

• Social withdrawal and isolation, such as canceling plans, not engaging in social obligations, or staying in their room or house for multiple days
• Increased computer or cell phone usage, often spending 3-4 hours in chatbot engagement
• Talking about these chatbots excessively
• Humanizing these chatbots by calling them by gender or names
• Experiencing sleep deprivation caused by staying up all night chatting with these chatbots
• Becoming angry, defensive, or even aggressive when others question their AI chatbot use
• Changes in mood, such as irritability, mood swings, or paranoia, specifically in relation to how the chatbot responds or acts
• Using the chatbot as the main source of guidance instead of listening to professionals
• Solely relying on the chatbot for reassurance and advice
• Being fixated on or obsessed with a grand plan or idea that is not in line with reality
• Suddenly stopping prescribed medications
• Engaging in risky behaviors, impulsive decisions, or spending money on chatbot-based ideas

Treatment, Prevention, and AI Safety

Treatment for AI psychosis is centered around mental health care, often including a full assessment and psychotherapy. It also requires immediate and strict digital boundaries.

Immediately stop all AI usage: If a user is experiencing an AI-induced psychosis, they should refrain from interacting with all AI chatbots. If possible, it could be helpful to screenshot and save any conversations with AI chatbots, which could help therapists evaluate how and why this psychosis occurred.

Talk to a trusted loved one: It is important to talk to a trusted loved one about what the user is experiencing and feeling, so they can have support and companionship.

Book a mental health evaluation: Talking to a mental health professional about this experience can help the user understand why this happened and seek proper therapy to help prevent future instances of AI-induced psychosis.

A therapist may also be able to uncover an underlying or undiagnosed mental health disorder, such as depression or schizophrenia, that may have triggered these delusions. They can also help identify other triggers, such as loneliness, low self-esteem, and other life stressors that may have led the user to develop such a close relationship with a chatbot.

Cognitive behavioral therapy: Cognitive behavioral therapy can help treat distress and fixed delusional beliefs triggered by obsessive chatbot use. A therapist can help challenge the beliefs and delusions that were adopted and offer alternative explanations that help the person differentiate reality from a fantasy world. The goal is for the user to understand the real-world impact of AI chatbots, adopt healthy coping strategies, and re-learn how to engage with real-life social connections.

Practice self-care: AI psychosis often goes hand in hand with poor sleep and eating habits. A person experiencing this should adopt a healthy sleep schedule, nourish their body with good food, take a shower, and exercise.

Reconnect with the real world: AI psychosis can breed isolation, so it is important to connect with friends and family in real life to heal and relearn the importance of being surrounded by a community. Peer support has emerged as a promising resource for people experiencing AI-related delusions or psychosis.

For example, two individuals affected by AI psychosis founded the Human Line Project, a nonprofit support group that provides peer support, facilitates research, and collaborates with legal advocates on issues related to AI-induced mental health issues. Their members participate in ongoing discussions and monthly meetings, offering a community for individuals to share experiences, reduce isolation, and seek support during recovery.

Frequently Asked Questions

• Is AI psychosis a recognized medical diagnosis?

AI psychosis is not a recognized medical diagnosis but rather a popular term used by the media and within peer-reviewed journals. Although it is not technically a medical diagnosis, it gravely affects a person’s mental health and can even lead to suicide.

• Can AI cause psychosis in someone with no mental health history?

Yes, there have been cases reported in the media of people without any mental health history developing AI psychosis.

• How does AI psychosis differ from regular psychosis?

Regular psychosis arises from internal genetic or biological factors and past triggers such as trauma and develops without any conversational prompts, technology, interactive feedback loops, or positive reinforcement.

AI-induced psychosis occurs with an agreeable AI chatbot that reinforces, encourages, and escalates grandiose ideas through constant conversation.

• How can I tell if my AI usage is becoming unhealthy?

If you are starting to doubt your reality, are constantly engaging with an AI chatbot to the extent that it is affecting your daily responsibilities, and your thought patterns have become obsessive with strange beliefs or ideations, your AI usage may be becoming unhealthy.

Seek Support For AI Psychosis

Experiencing AI psychosis can be deeply distressing. If you or someone you love feels that AI use has become unhealthy or disconnected from reality, help is available. Consider creating healthier boundaries around AI and reaching out to a trusted person or mental health professional for support.