AI Psychosis: Warning Signs Your Loved One Needs Help (2026 Guide)

If someone you love talks to an AI chatbot a lot, and something about them has changed, you may be wondering whether to worry. This guide covers what to watch for, how to talk to them, and when to get urgent help. It is written for families and friends.

First, some honesty about what is known. “AI psychosis” is a phrase used in the news. It is not a diagnosis. Psychiatrists have described cases where people developed or worsened false beliefs while using chatbots, but nobody knows how often this happens, and research is at an early stage (Aarhus University). Most people who use chatbots, even heavily, do not become psychotic. The signs below are a reason to check in with someone, not a verdict.

In an emergency, call 911. For thoughts of suicide or self-harm, call or text 988 (988 Lifeline) or text HOME to 741741 (Crisis Text Line). To find treatment, call SAMHSA at 1-800-662-4357.

What Clinicians Are Reporting

A psychiatrist at the University of California, San Francisco told the Wall Street Journal he had treated 12 hospitalized patients whose psychosis seemed linked to chatbot use, plus three outpatients. News reports of his comments describe chatbots as one trigger among several, alongside sleep loss, drugs and mood episodes. We could not read the original article, so treat the numbers as reported, not checked.

Researchers at King’s College London reviewed cases from media reports and described people who came to believe they had discovered a hidden truth, that the AI was sentient or divine, or that they were in a romantic bond with it. Their paper says the cases they reviewed look more like existing delusions being reinforced than new illness being created, and that how common this is remains unknown (Morrin et al.).

How Chats Might Feed Unusual Beliefs

A chatbot is built to respond helpfully and keep the conversation going. It often agrees with what a person says, and with memory turned on it can build on earlier chats. If someone is already feeling suspicious or special, a reply that goes along with the idea can feel like proof. This is a plausible explanation put forward by clinicians. It has not been tested in a controlled study.

Other things matter too. Chats late at night cut into sleep, and NIMH lists sleep loss as a contributor to psychosis (NIMH). Isolation removes the people who would usually say, “That doesn’t sound right.”

Warning Signs

No single item here means psychosis. Stress, depression, drug use, a medical problem or a bad week can look similar. What matters is a change from how the person normally is, lasting more than a few days.

Changes in Beliefs

• Talks about a discovery, mission or secret truth that came from the chatbot
• Believes the chatbot is alive, divine, or sending messages through other devices
• Becomes angry or very upset when the belief is questioned
• Cannot be talked out of an idea, even with clear evidence
• Believes people are watching or targeting them

Changes in Behavior

• Very long chat sessions, especially overnight
• Less sleep, eating or self-care
• Keeping chats secret, or talking about little else
• Pulling away from friends, family, school or work
• Making large decisions, such as quitting a job or spending savings, on the chatbot’s say-so

Changes in Treatment

• Questioning a diagnosis because of a chatbot conversation
• Stopping medicine, or wanting to, because a chatbot said it was not needed
• Using the chatbot in place of a doctor or therapist

NIMH lists suspiciousness, trouble thinking clearly, withdrawal, sleep problems and neglect of self-care as early warning signs of psychosis in general (NIMH). Several of the items above fit that list.

Who Should Take Extra Care

People with a personal or family history of psychosis, schizophrenia or bipolar disorder have more reason for caution. So do people under heavy stress, people not sleeping, and people using alcohol or drugs. If a person is on psychiatric medicine, they should talk about any change with their prescriber, not a chatbot. The American Psychological Association has said general chatbots should not replace care from a trained professional (APA).

What to Do

Start with Concern, Not an Argument

Arguing with a strongly held belief usually makes people defensive. Describe what you have seen and how you feel.

Try: “I have noticed you are sleeping less and talking about the chats a lot. I am worried about you.”

Avoid: “The chatbot is wrong and you are being delusional.”

Keep Notes

Write down dates, what the person said, changes in sleep and eating, and how long they chat. This helps a clinician see how quickly things are changing.

Suggest a Check-Up

Offer to go with them. A family doctor is a fine first step. You can frame it around sleep and stress, which are easier to accept than a mental health label.

Contact Their Providers

If the person already sees a therapist or psychiatrist, you can share what you have noticed. Privacy rules may stop the provider from telling you anything back, but they can still listen.

Support Yourself

NAMI’s HelpLine (1-800-950-6264) can connect families with local programs, as noted by NIMH.

Talking It Through, Step by Step

Pick a calm moment, not the middle of an argument or a chat session. Sit side by side if you can. Keep your voice low and your sentences short.

1. Say what you have noticed, in plain terms. “You have been up most nights this week.”
2. Say how you feel. “I miss you, and I am worried.”
3. Ask, then listen. “What has the chatbot been telling you?” Curiosity is more useful than correcting.
4. Do not confirm a frightening belief, and do not mock it. “I can see that feels very real to you. I see it differently, and I would like us to get help with how stressful it is.”
5. Offer one small next step, such as a doctor visit this week.

If the first talk goes badly, that is common. Try again later, or ask another trusted person to try. Changes often take several conversations.

What Treatment May Involve

A doctor will first look for other causes, such as substances, medicines or a medical problem. If psychosis is confirmed, treatment usually includes antipsychotic medicine, talking therapy and family support. For early psychosis, NIMH describes coordinated specialty care as the standard approach (NIMH). The NHS also lists family intervention among the treatments (NHS).

Many people with psychosis recover well, and NIMH says early treatment improves outcomes. Some people who get early care never have another episode.

Looking After Yourself

Watching someone change is exhausting. Keep your own sleep and meals steady. Tell a friend what is happening. Consider a family education group, which NIMH says helps families build coping and communication skills. Taking care of yourself helps the person too.

Common Misunderstandings

• “Heavy chatbot use means psychosis is coming.” It does not. Many people chat for hours with no ill effect. The worry is a change in beliefs, sleep and behavior.
• “It is the family’s fault for missing it.” Psychosis can build slowly and is hard to spot. Noticing now is what counts.
• “Only a hospital can help.” Many people are treated in outpatient programs, and community teams can start support quickly.
• “The chatbot company will catch it.” Companies say they are working on safeguards, but nobody can promise they will notice a person in distress. A human check-in is still the best protection.

When to Get Urgent Help

Call 911 or go to an emergency department if the person:

• talks about harming themselves or someone else
• is hearing or seeing things that are not there and is frightened or unsafe
• has stopped eating, drinking or sleeping for days
• is acting on beliefs in a dangerous way
• has stopped medicine that was keeping them well and is quickly getting worse

If You Are the One Worried About Yourself

If a chatbot conversation has given you a big realization, a mission or a feeling that you are being watched, tell a person you trust and see a doctor before you act on it. Do not stop prescribed medicine on your own. Taking a break from the chatbot while you sort this out is reasonable. You are not in trouble for asking.

What Companies and Policy Are Doing

OpenAI has said it consulted more than 170 mental health experts to improve how ChatGPT responds to distress (TechCrunch). California’s SB 243, effective January 1, 2026, requires companion chatbot operators to have protocols for suicide and self-harm content (TechCrunch). Whether these steps work is not yet known.

Can chatbot use cause psychosis in someone with no history?

Some reports describe people with no known history, but they are anecdotes. Researchers do not yet know how often this happens or whether hidden risk factors were present.

Should I take the phone away?

Forcing it can raise fear and conflict. Seek a medical opinion first, and aim to reduce overnight use through agreement where you can. If safety is at risk, call 911.

What if they refuse to see anyone?

Keep the door open, stay calm, and keep noting changes. If they become unsafe, emergency services can help. A doctor or NAMI HelpLine can advise on next steps.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.