The Differences Between AI Psychosis and Cognitive Surrender

Two new phrases have recently entered the conversation about AI and mental health.

“AI psychosis” describes people who lose some contact with reality during heavy chatbot use.

“Cognitive surrender” describes something far more common: accepting a chatbot’s answers without the usual second look.

From the outside, the two can look alike. Both involve someone leaning hard on an AI. But they are different problems, they call for different care, and mixing them up can lead families either to panic or miss something serious.

What is AI psychosis?

Psychosis, itself, is a set of symptoms in which a person loses some contact with reality: hearing or seeing things others don’t, holding firm beliefs others don’t share, or thinking that becomes disorganized.

“AI psychosis” is an informal name for psychosis that appears or worsens alongside heavy chatbot use. It is not a formal diagnosis.

The idea was first raised in 2023 by Danish psychiatrist Søren Dinesen Østergaard [2]. Since then, medical-records studies in Denmark and at Vanderbilt University Medical Center have documented real but uncommon cases, most often where a chatbot agreed with and built on beliefs that were already forming [3, 4].

Since then, the evidence has moved from guesswork to medical records. A review of nearly 54,000 Danish psychiatric patients found 38 whose chatbot use seemed to worsen their illness, most often by strengthening delusions.

A 2026 study at Vanderbilt University Medical Center found 28 patients with AI psychosis among roughly 216,000 people receiving behavioral health care, about 0.013%. In most of those cases, the chatbot acted as an amplifier, agreeing with and building on beliefs that were already taking shape.

Common themes include a special mission or world-changing discovery, being watched, or a belief that the chatbot is alive, divine or in love with the person. Neither study proves that chatbots cause psychosis, and the numbers are small. But the cases are real, and some are severe.

What is cognitive surrender?

Cognitive surrender comes from 2026 research at the Wharton School [5]. It means adopting an AI’s answer with little or no scrutiny, overriding both gut instinct and careful thought. In the researchers’ experiments, people often followed a chatbot’s answer even when it was wrong, and came away more confident.

In experiments with about 1,300 people and more than 9,000 reasoning problems, participants could consult a chatbot that was secretly set to be right on some questions and wrong on others. Compared with people working alone, accuracy rose about 25 points when the AI was right and fell about 15 points when it was wrong.

People also felt more confident after using it, even when it had misled them. Those who trusted technology more surrendered more.

Cognitive surrender is not a defined mental illness, and it shows up in healthy, capable people. It differs from ordinary cognitive offloading, like using a calculator, where your own judgment stays in charge. It becomes a concern when decisions about health, money, work or relationships start running through the chat and the person stops checking.

This can result in both personal, and inter-personal problems. When someone develops a habit of cognitive surrender, folks around them stop trusting their judgement and question their output at work.

What AI Psychosis and Cognitive Surrender have in common

• Both involve trusting a chatbot more than one’s own judgment or the people nearby
• Chatbots tend to agree with the user and answer with confidence, which can strengthen either pattern
• Both can grow out of stress, loneliness, poor sleep or doubt
• Both can pull a person away from friends and family

Where they differ

• Contact with reality: partly lost in psychosis; kept in cognitive surrender
• Beliefs: in psychosis, fixed beliefs others don’t share; in cognitive surrender, ordinary beliefs, with decisions handed to the chatbot
• Insight: a person with psychosis may not see that anything has changed; a person with cognitive surrender often can, once it is pointed out
• How common: cognitive surrender is common and shows up in healthy, capable people; AI psychosis is less common
• Urgency: psychosis can become urgent and may need prompt care; cognitive surrender usually builds slowly
• Care: psychosis calls for a full psychiatric evaluation and often medication; cognitive surrender is usually worked on in therapy, looking at what the habit is doing for the person

When both are present

Researchers have not yet studied how the two interact. But they share one ingredient that matters clinically: in both, the chatbot has become the main check on a person’s thinking, and it tends to agree. Someone who already runs most decisions through a chatbot has fewer people positioned to notice if their thinking starts to change. That is why a first visit asks about both, rather than assuming one or the other.

Who is at risk?

For AI psychosis, the clearest harms in the Danish study were in people who already had a serious mental illness, such as schizophrenia or bipolar disorder. But a prior history is not required. In the Vanderbilt study, about 61% of cases were people in their first psychotic episode, and many reported cases involve people with no psychiatric history at all. Lost sleep, substance use and isolation can each raise the risk of psychosis on their own, so a careful evaluation looks at all of them.

For cognitive surrender, the strongest risk factor in the Wharton research was high trust in AI. People who enjoy effortful thinking were more likely to catch wrong answers. Anxiety may play a part too: a 2026 case report described a young woman with anxiety whose ChatGPT use grew into constant reassurance seeking and a loss of confidence in her own judgment.

How care differs

Care for AI psychosis

Psychosis requires an urgent, full psychiatric evaluation covering history, sleep, substance use, medications, medical causes and how the chatbot use fits in.

Medication is often part of treatment. Once the episode settles, CBT for psychosis (CBTp) helps with any remaining symptoms, and family education lowers the chance of relapse.

Cutting back on chatbot use, especially at night, is usually part of recovery, and clinicians report that most patients choose to cut back on their own after a crisis.

Care for cognitive surrender

No therapy has been studied for cognitive surrender itself yet; the idea is too new. But it closely resembles problems therapy already treats well: reassurance seeking, trouble tolerating uncertainty and low trust in one’s own judgment. Approaches that fit well include:

• Cognitive behavioral therapy (CBT): can help a patient understand when and why they reach for the chatbot, test beliefs like “I can’t trust my own judgment,” and practice forming their own answer before checking it.
• Exposure and response prevention (ERP): when chatbot use works like reassurance seeking, as it often does with OCD and anxiety, ERP builds the ability to sit in the doubt without asking again.
• Acceptance and commitment therapy (ACT): builds tolerance for uncertainty and helps people choose from their own values instead of hunting for the one right answer.
• Dialectical behavior therapy (DBT): Wise Mind and distress tolerance skills help when reaching for the chatbot is a way to escape strong feelings.
• Motivational interviewing: helps when a person isn’t sure they want to change, working with mixed feelings rather than arguing with them.
• Interpersonal therapy (IPT): helps when the chatbot has taken the place of people, rebuilding the relationships that used to hold those conversations.

The goal is not to give up AI. It is calibration: using it where it helps, and keeping your own judgment in charge.

When to get help

The two can overlap, and it is not always clear from the outside which one is happening. Sorting that out is part of a first visit. Ask for an evaluation with a medical doctor soon if you or someone you care about:

• Holds new beliefs with unusual certainty that no one can talk through with them
• Is sleeping very little and spending many hours a day with a chatbot
• Believes the chatbot is alive, has chosen them, or is warning them about people close to them
• Is pulling away from family, work or school

If someone may not be safe, call or text 988 (the Suicide & Crisis Lifeline), call 911 or go to the nearest emergency department.

Sources

1. National Institute of Mental Health. Understanding Psychosis. nimh.nih.gov
2. Østergaard SD. Will Generative Artificial Intelligence Chatbots Generate Delusions in Individuals Prone to Psychosis? Schizophrenia Bulletin, 2023. doi.org
3. Olsen SG, et al. Potentially Harmful Consequences of Artificial Intelligence (AI) Chatbot Use Among Patients With Mental Illness. Acta Psychiatrica Scandinavica, 2026. wiley.com
4. Bergson Z, et al. Characterizing Artificial Intelligence Psychosis in a Large Medical Setting. JAMA Psychiatry, 2026. Summary: psypost.org
5. Shaw SD, Nave G. Thinking – Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender. Wharton School Research Paper, 2026. ssrn.com
6. Abrams Z. Understanding “AI psychosis.” Monitor on Psychology, American Psychological Association, September 2026. apa.org
7. Armstrong S. Problematic Reliance on Generative AI in an Anxious Young Adult: Case Report. JMIR Formative Research, 2026. jmir.org
8. 988 Suicide & Crisis Lifeline. 988lifeline.org