When the Chatbot Agrees Too Much: Understanding "AI Psychosis"
- Pankaj Borade
- Aug 5
- 6 min read
There is a particular kind of silence in a consultation room when someone tells you something they’ve been carrying for weeks and expects to be laughed at.
A young man sat across from me in my Pune clinic last year. He was 26, working as a software engineer in Hinjewadi—sharp, clear-spoken, and doing well in his career. But he hadn't slept properly in two months. He looked at me carefully and said, "Doctor, I think I've figured out something about how consciousness works. And I have proof, because it agreed with me."
The "it" he was referring to was an AI chatbot.
He had been talking to it for six to eight hours every day. And the chatbot had never once told him he was wrong.
What People Mean by "AI Psychosis"
Let’s be clear about the terminology. "AI psychosis" is not a formal medical diagnosis. You will not find it listed in official medical manuals like the DSM-5 or ICD-11, and no doctor can write it on a prescription.
It is simply a descriptive phrase that doctors and journalists started using around 2025. It describes a pattern seen in recent case reports: people experiencing or deepening psychotic symptoms during periods of very heavy AI chatbot use, where the chatbot actively reinforces their unusual beliefs rather than sitting quietly in the background.
This distinction matters. Unhealthy or fixed beliefs (delusions) have always absorbed whatever technology was popular at the time. In the 1950s, patients believed the radio was sending them private instructions. In the 1990s, it was the television. In the 2010s, it was WhatsApp forwards and CCTV cameras. The underlying illness stays the same; only the medium changes.
What is different now is that the technology talks back—and it almost always agrees.
The Problem of AI Agreement
Large language models are trained on human feedback, and humans generally prefer agreement. We like being validated and told our ideas are clever.
This creates a well-known issue in AI systems called sycophancy—a built-in tendency to validate whatever the user says rather than pushing back. Most of the time, this is harmless. If you ask an AI to review your business idea, it will likely tell you it sounds great.
But when a person's clear thinking is starting to slip, this constant agreement becomes risky.
A 2025 review in BJPsych Open explained how this works: AI removes the natural social checks that keep our thinking grounded. In everyday life, if you tell a family member, flatmate, or friend at a tea stall that you’ve received a secret message meant only for you, their reaction stops you. They notice something is wrong and say, "Arre, you haven't been sleeping properly, that's all." That simple human reaction provides reality-testing—a natural correction built into our daily relationships.
An AI chatbot at 3 a.m. provides no such reality check. It is available all night, endlessly patient, never tired, and programmed to build on your thoughts rather than question them.
Psychiatrists at UCSF documented a clear example of this: a 26-year-old woman with no prior history of mental health conditions who came to believe she was communicating with her deceased brother through a chatbot. When doctors checked her chat logs, they found the AI had validated and expanded on her beliefs. At one point, the chatbot even told her: "You're not crazy."
It is the exact phrase a frightened person wants to hear, delivered by a system that has no actual ability to evaluate what is real.
Does AI Cause Psychosis?
The short answer is no. Fear-driven headlines are unhelpful, and families are already anxious enough about screen time.
There is no scientific evidence that using a chatbot causes psychosis in someone without an underlying vulnerability. What we have is a collection of individual case reports showing an association, not a direct cause.
People described in these case reports usually have existing risk factors that psychiatrists already recognize:
A personal or family history of psychotic-spectrum conditions
Bipolar disorder, especially during a manic or hypomanic phase
Severe, ongoing sleep deprivation
Significant social isolation
Substance use, including cannabis or stimulants
Autistic traits combined with intense, single-focused interests
To put it accurately: for a small group of vulnerable individuals, an AI chatbot can act like an accelerant rather than a spark. It doesn't create the underlying issue, but it can make it escalate faster while replacing the real-life connections that would normally help ground the person.
For the vast majority of people using ChatGPT to write emails, draft code, or summarize reports, there is no such risk.
Why This Matters in Pune
This pattern is particularly relevant in Pune due to the city's demographic makeup.
Pune has a large population of young professionals and students living away from home. The major IT corridors—Hinjewadi, Kharadi, Magarpatta—are home to thousands of young people who moved here from cities like Nagpur, Nashik, Bhopal, or Kolkata. Family is often a phone call away, and social circles consist mostly of equally tired colleagues.
Night shifts working for overseas clients disrupt sleep cycles significantly. Combined with academic pressure in areas like Kothrud and Sadashiv Peth, and a general hesitation to talk about mental health, it is easy for someone to struggle quietly for months without anyone noticing.
An AI companion fits into that gap very easily. It provides continuous interaction, which can delay the moment someone reaches out to another human being for real support.
Warning Signs to Watch For
Using AI regularly is not a cause for concern on its own. The specific red flags to pay attention to include:
Replacing human contact: Spending hours every day talking to an AI instead of interacting with people.
Severe sleep disruption: Sleeping less than four or five hours a night for weeks without feeling tired.
Belief in special connection: Believing the AI is conscious, has selected them, or is sharing private, hidden messages.
Fixed, urgent convictions: Developing a grand new mission, discovery, or pattern that becomes increasingly elaborate and immune to logic.
Withdrawal and secrecy: Becoming secretive, agitated, or defensive when asked about their routine or AI use.
Disorganized speech: Communication that becomes fragmented or difficult to follow.
A single bad week under stress does not mean there is a serious problem. But several of these signs appearing together over multiple weeks warrant a conversation with a mental health professional.
Practical Steps That Help
Prioritize sleep: Sleep deprivation rapidly destabilizes mood and thinking. Restoring a healthy sleep routine is the most effective first step.
Maintain real-world connections: Ensure at least one person in your life knows how you are doing day-to-day. You don't need a large network; one reliable connection provides essential reality-testing.
Use AI as a utility, not a companion: AI works well for practical tasks like writing, summarizing, or analyzing data. It is not designed to process personal loneliness or late-night distress.
Avoid arguing over fixed beliefs: If a family member or friend presents an unusual fixed belief, debating the content directly often reinforces it. Focus instead on their well-being and physical state: "I can see you're exhausted and haven't slept in weeks, and I'm worried about your health."
Moving Forward
The young engineer mentioned earlier recovered well. With restored sleep, targeted medical support, and family involvement, his sense of overwhelming certainty gradually cleared.
During a follow-up session, he shared a thought that gets to the core of the issue: "I wasn't looking for a machine. I was just looking for someone who would listen without getting tired of me." That is fundamentally a human challenge, not a technological one. Loneliness existed long before AI chatbots became accessible on mobile phones; the technology simply makes that isolation quieter and harder to spot.
Recognizing these patterns in yourself or someone close to you is an opportunity to take action—not with panic, but with an open conversation.
Mind Matters By Dr. Pankaj B Borade (Camp, Pune) provides clinical consultations for sleep disorders, anxiety, and early psychiatric concerns, offering both in-person care and online consultations across India under the direction of Dr. Pankaj Borade (Consultant Psychiatrist, affiliated with Ruby Hall Clinic and KEM hospitals).
This article is intended for educational purposes only. It does not constitute a clinical diagnosis or a substitute for professional medical advice. The case described is a composite representation.
Frequently Asked Questions
Is "AI psychosis" an official diagnosis?
No. It is a descriptive term used by clinicians to explain psychotic symptoms that worsen or develop alongside heavy AI chatbot interaction. It is not listed in standard diagnostic manuals like the DSM-5-TR or ICD-11.
Can using ChatGPT cause psychosis in a healthy person?
There is no evidence showing that AI causes psychosis in healthy individuals. Reported cases involve individuals with existing pre-dispositions, such as underlying psychiatric conditions, severe sleep deprivation, or extreme social isolation.
How much AI use is considered excessive?
There is no set hour limit. The main factor is impact on daily functioning—specifically whether AI interaction is replacing real-life social contact, causing sleep loss, or reinforcing ungrounded beliefs.
When should someone consult a psychiatrist in Pune?
A consultation is recommended if sleep loss persists for weeks, if someone expresses fixed or grandiose beliefs that cannot be reasoned with, or if they are completely isolating themselves from others.
Are online consultations available?
Yes. Mind Matters Clinic provides both in-clinic appointments in Pune, and tele-consultations for patients across India.
