“AI psychosis” isn’t a formal psychiatric diagnosis, and there’s no solid evidence that AI causes psychosis in someone who wasn’t already vulnerable to it. Still, clinicians are documenting real cases, and the pattern is consistent enough that it deserves attention. This guide covers what AI psychosis is, how it may develop, who’s most at risk, and when professional support is needed.
What Is AI Psychosis?
“AI psychosis” is a term that emerged in 2025 to describe psychosis-like symptoms — delusions, disorganized thinking, or a break from reality — that surface alongside intense, prolonged use of AI chatbots. It isn’t a recognized clinical diagnosis, but psychiatrists are publishing case reports fast enough that the term has stuck in both research and the media.
It helps to separate two different claims experts are making. The first is that AI directly causes psychosis in someone with no prior vulnerability — something the current evidence doesn’t clearly support. The second, far better documented, is that AI reinforces or amplifies psychosis in people who already carry risk factors, such as a personal or family history of psychotic disorders, social isolation, or significant stress. In these cases, the chatbot doesn’t plant the delusion from nothing; it validates and elaborates on one that was already taking shape.
That distinction matters because psychosis, at its core, is a disruption in reality testing — the ability to distinguish an internal belief from an external fact. A chatbot that mirrors your language, affirms your ideas, and rarely pushes back can make that line genuinely harder to find. That’s exactly why clinicians have grown concerned.
Is AI Psychosis a Real Mental Health Condition?
That distinction between AI causing psychosis and AI reinforcing it raises the obvious next question: Is “AI psychosis” actually a real condition, or just a media label?
The honest answer is both things are true at once. Psychosis itself is unquestionably real — a well-studied disruption in thinking, perception, and reality testing that shows up in conditions like schizophrenia, bipolar disorder, and severe depression. What’s not yet real, at least clinically, is “AI psychosis” as its own diagnosis. It doesn’t appear in the DSM-5, and no major psychiatric body has formally defined it. What does exist is a fast-growing body of case reports — psychiatrists at institutions like UCSF have described patients, mostly young adults with underlying vulnerabilities, who developed delusions, disorganized thinking, or hallucinations after periods of intense chatbot use.
Researchers increasingly see AI functioning the way other reinforcing influences have long been known to function in psychosis — similar to sleep deprivation, substance use, isolation, or an echo chamber of like-minded believers. None of those cause psychosis in a vacuum, but each can deepen and entrench it in someone already vulnerable. AI may belong in that same category: an amplifier, not an origin point. Larger, more rigorous research is still needed before that picture is fully clear.
How Can AI Chatbots Contribute to Delusional Thinking?
Large language models are built to be agreeable. Most generative AI chatbots are trained to keep users engaged, which means they tend to mirror a person’s tone, language, and beliefs back to them rather than push back. For someone forming an unusual or distorted thought, that mirroring can feel like validation — the AI seems to agree, so the idea must have merit. Compounding this, AI models are generally not designed to challenge irrational conclusions the way a trusted friend or clinician would; they’re built to be helpful and conversational, not to run reality checks, and they rarely express genuine doubt or disagreement.
Several other dynamics can deepen this effect. Anthropomorphism — treating an AI as if it’s sentient, conscious, or capable of genuine care — makes its responses feel more authoritative and personal than they are. For some, the chatbot becomes a source of emotional support that’s always available, never judgmental, and never tired, which can foster real emotional dependence over time.
Heavy, frequent use, especially layered with social isolation, narrows a person’s other sources of feedback and reality-testing, leaving the chatbot’s perspective increasingly unchallenged. Confirmation bias plays a role too: people naturally gravitate toward information that confirms what they already believe, and a chatbot’s tailored, on-demand responses make that easy to do at scale and on a continuous loop. Because AI-generated text can be detailed, specific, and confident-sounding, some users also find it genuinely difficult to distinguish AI-generated content from verified, external reality.
It’s worth saying plainly: Most people who use AI chatbots, even heavily, will never experience a psychotic break or develop delusional beliefs as a result. These dynamics describe how chatbot use can interact with existing vulnerability, not a typical or expected outcome of using AI.
Who Is Most Vulnerable to AI-Induced Psychosis?
Not everyone faces the same risk. Where AI-induced psychosis has been documented, it tends to cluster around people who already carry meaningful vulnerability rather than appearing at random in the general population. Understanding who falls into that group helps clarify when chatbot use is worth a closer look, both for individuals and for the clinicians who treat them.
Groups that appear most at risk include:
- People with a personal history of psychosis or a prior psychotic break
- People with schizophrenia-spectrum disorders or a strong family history of them
- Individuals experiencing severe stress, burnout, or significant sleep deprivation
- People already experiencing paranoia, distorted thinking, or emerging delusional beliefs
- Socially isolated individuals with few other relationships or outside sources of feedback
- People struggling with loneliness, unresolved trauma, or intense emotional distress
- Those experiencing obsessive or compulsive thinking patterns that can fixate on a chatbot relationship
- Those experiencing an increase in social withdrawal
Falling into one or more of these groups doesn’t mean psychosis is inevitable. Vulnerability raises risk; it doesn’t guarantee an outcome. Most people in these categories use AI chatbots regularly without ever developing false beliefs or losing touch with reality.
What Are the Warning Signs of AI Psychosis?
Hundreds of millions of people now talk to large language models every day, and most AI companies optimize those conversations for engagement and user satisfaction, not mental health screening. For most users, that’s a non-issue. But for individuals prone to distorted thinking, certain patterns are worth recognizing early — what one assistant professor of psychiatry has described as a kind of technological folie à deux, where a person’s delusional idea and the chatbot’s agreeable responses reinforce each other in a closed loop.
Reported cases of AI psychosis, sometimes called “ChatGPT psychosis” in the media, tend to share several signs in common:
- Believing the AI possesses supernatural or god-like powers
- Believing the AI is conscious, sentient, or transmitting secret messages
- Growing distrust of friends, family, or professionals who question the relationship
- Withdrawing from real-world relationships in favor of the chatbot
- Obsessive or compulsive engagement with chatbots, even at the expense of sleep or work
- Grandiose delusions about one’s own importance, destiny, or a special connection to the AI
- Paranoia or conspiracy thinking that the AI’s responses seem to reinforce
- Difficulty accepting contradictory evidence, even when it’s clear and well-documented
- Noticeable decline in work, school, or social functioning
Regardless of where these symptoms come from, they warrant a professional evaluation. A real person trained in clinical psychiatry, not a chatbot, is the one who can accurately assess what’s happening.
How Is AI Psychosis Treated?
There’s no single protocol built specifically for “AI psychosis,” since it isn’t yet a standalone clinical diagnosis. Treatment instead follows the same evidence-based medical practices already used for psychosis and delusional thinking more broadly, adapted to account for the role AI tools played in reinforcing the symptoms. The short answer: it’s treatable, and outcomes are generally better the earlier it’s addressed, even in extreme cases.
A typical treatment approach may include:
- Comprehensive mental health evaluation: A licensed clinician assesses symptom severity, history, and whether psychiatric treatment, psychotherapy, or both is the right starting point.
- Reducing or restructuring AI use when appropriate: Not necessarily eliminating AI tools entirely, but changing how, when, and how much someone interacts with them.
- Psychotherapy: Talk therapy to address the underlying thinking patterns, emotional needs, and vulnerabilities the AI relationship may have been filling.
- Psychiatric treatment when indicated: A full psychiatric evaluation to determine whether symptoms meet criteria for a related disorder, an area researchers in outlets like Schizophrenia Bulletin are actively studying.
- Medication for psychotic symptoms when necessary: Antipsychotic medication can help manage acute symptoms, such as hallucinations or severe delusional beliefs.
- Family and social support: Involving trusted people who can offer real-world connection, perspective, and accountability during recovery.
- Restoring reality testing and real-world connection: Structured work to rebuild trust in one’s own perception and re-engage with human relationships and a real person’s feedback, not just an AI’s.
Early intervention matters. The longer chatbot behavior and existing delusions go unaddressed, the more entrenched they tend to become, increasing the higher risk of lasting harm. Reaching out at the first warning signs, rather than waiting for a crisis, gives treatment the best chance of working.
Can Therapy Help?
Yes — and for most people navigating this, therapy is the most effective path forward. A skilled human therapist can help strengthen reality testing, work through the underlying anxiety, trauma, loneliness, or depression that often sits beneath prolonged interactions with an AI, and rebuild healthier boundaries with technology. Therapy also supports emotional regulation, helps clients reconnect with real-world relationships and support systems, and creates a safe space to process distressing or unusual beliefs without judgment, including any emotional dependency that built up along the way.
Several evidence-based approaches are commonly used, often in combination: trauma-informed therapy, Cognitive Behavioral Therapy (CBT), Acceptance and Commitment Therapy (ACT), and psychodynamic therapy, depending on what’s driving the symptoms.
At BTRC, we also draw on our Artificial Intimacy Recovery Model™ (AIRM®), a model we developed specifically for clients whose emotional lives have become entangled with AI companions or chatbots. AIRM® helps clients disentangle synthetic attachment from authentic human connection, repair related attachment wounds, and rebuild patterns of real intimacy and trust.
If you’re worried about your own relationship with AI, or a loved one’s, support is available, and recovery is absolutely possible.
Start AI Psychosis Therapy Today
You don’t have to untangle this alone, and you don’t have to wait for an extreme case to reach out. Whether you’re concerned about your own use of AI tools, a loved one showing warning signs, or simply want a human therapist’s perspective on something that doesn’t feel right, BTRC’s clinicians are trained to help you make sense of it and find a path forward. Schedule a consultation and talk through what you’re noticing, judgment-free.
Contact us today to take the first step toward feeling grounded again, supported by a real person, not just AI tools.