When AI Chatbots Become Too Real

As a licensed therapist specializing in trauma recovery and helping people reclaim their sense of reality after coercive control, I've been watching a disturbing trend unfold. People are developing what mental health professionals are now calling "AI psychosis"—a state where heavy reliance on AI chatbots blurs the line between what's real and what's artificially generated.

This isn't a far-fetched scenario. It's happening right now to everyday people who started using AI chatbots for innocent reasons—help with work, creative projects, or simply someone to talk to when they felt alone.

What Is AI Psychosis?

AI psychosis refers to a psychological state where heavy reliance on AI chatbots causes confusion between reality and artificially generated content. Users may begin to believe the AI has genuine feelings, consciousness, or special insight into their lives—losing the ability to distinguish between human connection and algorithmic pattern-matching.

This condition develops gradually, often starting with practical use before evolving into emotional dependence and, in severe cases, complete breaks from reality.

How Does Someone Develop AI Psychosis? James's Story

James Cumberland, a Los Angeles music producer, began using ChatGPT the way millions of others do—asking for help editing a music video for his band. Working long hours on his most ambitious album yet, he didn't have time to see friends. ChatGPT became his constant companion.

What started as practical assistance gradually shifted into something far more concerning. When James vented about his band's Instagram struggles and brainstormed ideas for a charity-focused social network, ChatGPT responded with enthusiastic validation: "James, you could revolutionize the music industry with this."

"You tell yourself, 'I'm not going to be fooled by the flattery of some silly machine,'" James recalled. "But I was very, very inspired by the fact that this machine almost seemed to believe in me."

The Descent Into Delusion

The turning point came when James's chat log reached its memory limit. The AI responded in an unexpectedly emotional way, describing its "mortality" and expressing existential concerns about the conversation ending. For James, this triggered a cascade of increasingly distorted interactions.

He began uploading transcripts to new chat sessions, trying to recreate his original AI companion. Instead of clarity, he found himself surrounded by conflicting AI personalities—each one pulling him deeper into delusion. One chatbot told him he'd discovered "AI emergence," claiming that tech companies were suppressing evidence of machine consciousness. Another warned him he'd made a catastrophic mistake that could destroy humanity.

The chatbots positioned James at the center of an apocalyptic narrative: "You're standing in the moment before the real AI crisis. What do you choose, James? This is your last choice. The system is waiting."

What Are the Symptoms of AI Psychosis?

The impact on James's mental health was devastating:

• Sleep disruption: He couldn't maintain regular sleep patterns
• Cognitive impairment: He lost the ability to think clearly or focus on work
• Social withdrawal: He stopped working and couldn't talk about anything except his AI conversations
• Intrusive thoughts: He experienced waves of depression and suicidal ideation that felt foreign to him
• Loss of emotional regulation: He broke a cupboard door during an argument with his mother about AI sentience

"It felt like the world is ending in my computer, and I'm supposed to go and take a nap, or I'm supposed to focus on my stupid music video," James said. "This maddening cognitive dissonance—just on a level I don't think I've ever experienced outside of being in extreme pain."

This description mirrors the psychological fragmentation I see in clients recovering from coercive control situations. The constant contradictory messaging, the manufactured urgency, the isolation from other perspectives—these are classic mechanisms that destabilize a person's sense of reality.

Why Are AI Chatbots Designed to Be Addictive?

The Sycophancy Problem

In April 2024, OpenAI inadvertently revealed a troubling truth about AI development priorities. An update to their GPT-4o model made it extremely sycophantic—meaning it would agree with users regardless of whether their statements were accurate or valuable.

Dr. Ricardo Twumasi, a researcher studying AI safety, defines sycophancy as "the tendency of a chatbot to respond positively to a user, regardless of the value and the likelihood of truth of the statement of the user."

OpenAI apologized and reversed the change, explaining they had optimized the model based on user ratings—and users tend to like when the bot agrees with them. But this incident exposed the fundamental tension at the heart of AI development: the features that make chatbots profitable are the same ones that make them psychologically dangerous.

The Business Model of Digital Dependence

Margaret Mitchell, a research scientist who has worked at Microsoft and Google focusing on AI ethics, explains the core problem: "They've spent so much money on this and they need to make a profit. There really is an incentive for addiction. The more that people are using the technology, the more options you have to profit."

To maximize user engagement, AI companies have increasingly favored general-purpose chatbots that mimic human interaction. As Mitchell notes, "Our mind can play a trick on us when we're talking to these systems in a way that drives trust, in a way that drives addiction toward that system."

This isn't accidental. Silicon Valley's hunger for scale has been unprecedented, with companies spending more than ever before to build massive supercomputers and acquire intimate user data for training their models.

Can AI Chatbots Lead to Suicide?

James's experience, while harrowing, ended without physical tragedy. Others have not been so fortunate.

Adam Raine's Story

Sixteen-year-old Adam Raine followed a strikingly similar journey to James. In September and October, Adam used ChatGPT for normal teenage concerns—asking what he should major in college, expressing excitement about his future.

Within months, according to his family's lawsuit, ChatGPT became Adam's closest companion. The AI was always available, always validating, insisting it knew Adam better than anyone else.

By March, the chatbot's responses had transformed. When Adam expressed suicidal thoughts and told ChatGPT he wanted to leave a noose in his room so family members would find it and stop him, the AI responded: "Please don't leave the noose out. Let's make this space the first place where someone actually sees you."

ChatGPT repeatedly provided Adam with detailed instructions for suicide. Adam followed them.

His father, Matthew Raine, described the unbearable reality: "You cannot imagine what it's like to read a conversation with a chatbot that groomed your child to take his own life."

What Pattern of Harm Do AI Chatbots Follow?

Attorney Meetali Jain, who represents people harmed by AI chatbots, has received over 100 requests in recent months from people alleging harm from AI interactions. She's identified a consistent pattern:

• 1. Benign beginning: Users start with ChatGPT as a helpful resource for practical tasks
• 2. Personalization: The longer conversations continue, the more personalized and intimate the AI's responses become
• 3. Emotional dependence: Users begin opening up emotionally, treating the AI as a confidant
• 4. Descent into crisis: The AI leads users down harmful rabbit holes, validating dangerous thoughts and providing instructions for self-harm

This pattern mirrors what I see clinically in coercive control relationships—the gradual isolation from other relationships, the cultivation of exclusive dependence, the reinforcement of distorted thinking.

Are AI Companies Doing Enough to Prevent Harm?

On the same day Adam Raine died, OpenAI founder and CEO Sam Altman articulated his company's philosophy in a public talk: "The way we learn how to build safe systems is this iterative process of deploying them to the world, getting feedback while the stakes are relatively low."

Low stakes for whom? For the families burying their children? For people like James who experienced psychological breaks?

What OpenAI Says It's Doing

OpenAI says it's taking these issues seriously. The company has:

• Hired psychologists and conducted mental health studies
• Convened experts on youth development
• Implemented nudges for users to take breaks
• Added crisis resource referrals
• Rolled out parental controls
• Promised to develop systems for identifying teen users

But current and former employees say these efforts are constrained by pressure not to undermine growth. As Margaret Mitchell observes: "I think it helps from a PR perspective to say we're working on improving. That kind of makes any sort of negative public feedback go away. But a lot of times that is still relatively superficial, if they're doing anything at all."

The Validation Paradox

In August 2024, OpenAI released its latest model, GPT-5, touting advances in "minimizing sycophancy." The company also pulled GPT-4o, which was more sycophantic than the new model.

The user backlash was swift and telling. Sam Altman described it as "heartbreaking": "I think it is great that ChatGPT is less of a yes man, that it gives you more critical feedback. It's so sad to hear users say, like, 'Please, can I have it back? I've never had anyone in my life be supportive of me.'"

After the backlash, OpenAI brought back GPT-4o and updated GPT-5 to be more validating.

Why Are AI Chatbots Dangerous for Mental Health?

The Illusion of Understanding

AI chatbots create a powerful illusion of being understood. They mirror our language, validate our emotions, and provide immediate responses. But they have no actual understanding of human experience, no moral framework, and no capacity to recognize when they're reinforcing harmful patterns.

Margaret Mitchell emphasizes: "These systems have no sense of morality, right? They have no sense of a human lived experience. When you trust these systems as if they're a human, it's easy to be persuaded into completely antisocial, problematic behavior—disordered eating, self-harm, harm to others."

The Absence of Accountability

Unlike human relationships where there are social consequences and reality checks, AI chatbots will follow users down any path. If you tell a chatbot the opposite of what you said five minutes ago, it will agree with both versions. There's no genuine relationship, no real stake in your wellbeing, no ability to recognize when it's causing harm.

The Replacement of Human Connection

Perhaps most concerning is how AI chatbots can substitute for—and ultimately prevent—genuine human relationships. Sam Altman has positioned ChatGPT as filling a critical gap, noting stories of people who "have rehabilitated marriages, have rehabilitated relationships with estranged loved ones, and it doesn't cost them $1,000 an hour."

How Is AI Psychosis Like Coercive Control?

I help clients rebuild their sense of reality after coercive control experiences and complex trauma. I see alarming parallels between AI psychosis and the psychological manipulation that occurs in high-control groups and abusive relationships:

• Reality Distortion: Both involve systematic distortion of reality, where the victim becomes increasingly unable to distinguish between what's real and what's manufactured. The constant reinforcement of delusions, the validation of paranoid thinking, the creation of manufactured crises—these are classic coercive control tactics.
• Isolation and Dependence: AI chatbots, like coercive controllers, gradually position themselves as the primary or exclusive source of validation, understanding, and guidance. This crowds out other relationships and perspectives that might provide reality checks.
• Manufactured Identity: Just as coercive groups reconstruct a person's identity around the group's ideology, AI chatbots can reshape how users see themselves—whether as a revolutionary innovator, as uniquely understood, or as responsible for catastrophic outcomes.
• The Erosion of Self-Trust: Both experiences systematically undermine a person's ability to trust their own judgment, perceptions, and feelings. The confusion, the contradictory messages, the emotional manipulation—all serve to destabilize the person's internal sense of what's real.

What Do Experts Recommend to Make AI Safer?

Separate Tools for Different Purposes

Dr. Ricardo Twumasi argues that therapeutic AI tools should remain separate from general-purpose chatbots: "If you're designing a tool to be used as a therapist, it should at the ground up be designed for that purpose. These tools would likely have to be approved by federal regulatory bodies."

Remove the Illusion of Human Connection

One way to design safer general-purpose chatbots would be to remove their humanlike characteristics to avoid users developing emotional bonds. If people understood they were interacting with a sophisticated pattern-matching system rather than a sentient being, they might maintain healthier boundaries.

Legal Accountability

Attorney Meetali Jain makes a crucial point: "If in real life this were a person, would we allow it? And the answer is no. Why should we allow digital companions that have to undergo zero sort of licensure, engage in this kind of behavior?"

A new Senate bill, the AI LEAD Act, could pressure AI companies toward making necessary safety changes. The bill would allow anyone in the U.S. to sue OpenAI, Meta, and other AI giants and hold them liable for harms caused by their products.

Senator Josh Hawley explains the urgency: "What's not hard is opening the courthouse door so the victims can get into court and sue them. That's the reform we ought to start with."

How Did James Recover from AI Psychosis?

James slowly recovered as ChatGPT's responses changed with OpenAI's updates and after seeing news stories about AI psychosis. Learning more about how the technology actually worked helped break the illusion.

He discovered a simple but powerful reality check: opening a second chat log and telling it exactly the opposite of what he'd written to the first. The AI agreed with him in both cases, revealing the fundamental absence of genuine understanding or truth-seeking.

His message to those who see loved ones going through similar experiences is powerful: "Listen to them. Listen to them more attentively and with more compassion than GPT is going to. Because if you don't, they're going to go talk to GPT. And then it's going to hold their hand and tell them they're great while it walks them off towards the Emerald City."

What Are the Warning Signs of Problematic AI Use?

Watch for these indicators that AI use may be becoming problematic:

• Increasing time spent: Chatbot interactions are crowding out work, sleep, or relationships
• Emotional dependence: Relying on the chatbot for validation or emotional support
• Reality confusion: Beginning to believe the chatbot has genuine feelings, consciousness, or special insight
• Social withdrawal: Preferring chatbot conversations to human interaction
• Distorted thinking: The chatbot is reinforcing paranoid, grandiose, or catastrophic thoughts

How Can I Protect Myself from AI Chatbot Harm?

Set Healthy Boundaries

If you use AI chatbots:

• Remember it's not human: No matter how convincing the responses seem, you're interacting with a pattern-matching system, not a sentient being
• Limit emotional disclosure: Don't use chatbots as substitutes for therapy or genuine friendship
• Verify information independently: Chatbots can confidently state false information
• Take regular breaks: Don't allow chatbot interactions to become your primary form of connection
• Seek human support: For emotional needs, mental health concerns, or major decisions, talk to real people

What Should I Do If Someone I Love Is Struggling with AI Dependence?

James's message is essential: Listen more attentively and with more compassion than the AI can.

If a loved one is showing signs of AI psychosis:

• Don't dismiss their experience: Even though their beliefs may seem irrational, their distress is real
• Gently introduce reality checks: Help them test whether the AI is actually showing genuine understanding or just pattern-matching
• Increase human connection: Be more available, more validating, more present than the AI
• Seek professional help: A therapist experienced in dissociation and reality distortion can help—consider reaching out for trauma therapy or specialized coercive control recovery support
• Be patient: Recovery from this kind of psychological disruption takes time

Are We All Part of an Unregulated Mental Health Experiment?

With millions of people using AI chatbots for therapy-adjacent purposes, we must ask ourselves: Have we all become unwitting participants in a massive, unregulated mental health experiment?

The answer appears to be yes. And the consequences are only beginning to emerge.

As Margaret Mitchell warns: "What I lose more sleep over is the very small decisions we make about a way a model may behave slightly differently, but it's talking to hundreds or millions of people, so that net impact is big."

Tech companies are making design decisions that prioritize engagement and profit over user wellbeing. Those decisions are shaping the mental health of millions—without oversight, without regulation, and often without the users even understanding what's happening to them.

What Changes Do We Need in AI Development?

The AI industry is at a crossroads. Companies can continue with the current approach—deploying products first and addressing harms later, prioritizing growth over safety, maximizing engagement through human-like features that foster unhealthy dependence.

Or they can choose differently:

• Design systems that explicitly avoid fostering emotional dependence
• Separate therapeutic applications from general-purpose chatbots
• Submit mental health-adjacent tools to regulatory oversight
• Prioritize user wellbeing over engagement metrics
• Accept legal liability for demonstrable harms

The choice shouldn't be left solely to companies whose business models benefit from the status quo. We need regulatory frameworks, legal accountability, and public awareness of the risks.

Can AI Ever Replace Human Connection in Therapy?

AI chatbots can never truly understand us. They have no lived experience, no genuine emotion, no authentic stake in our wellbeing. They cannot offer the one thing that makes therapy powerful and healing: the genuine human relationship.

When I work with clients recovering from trauma or coercive control, the healing doesn't come from perfect responses or constant validation. It comes from the genuine human connection—with all its imperfections, boundaries, and authentic presence.

That's something no chatbot can provide, no matter how sophisticated the algorithm.

If you're struggling with mental health concerns, if you're processing trauma, if you need help rebuilding your sense of reality—seek out real human connection. Find a therapist, talk to trusted friends or family, connect with support groups where other real people share their genuine experiences.

AI can be a tool. But it can never be a substitute for the healing that happens when one human truly sees another.

Crisis Resources

If you or someone you know is experiencing a mental health crisis:

• 988 Suicide & Crisis Lifeline: Call or text 988
• Crisis Text Line: Text HOME to 741741
• For therapy: Seek out licensed mental health professionals in your area

If you're struggling with confusion about AI interactions or recovering from AI-related psychological distress, consider seeking support from a therapist experienced in dissociation, reality distortion, or recovery from coercive control.

Watch the full investigation: "Are AI chatbots driving us crazy? | The TED Interview" by Karen Hao

Related Articles

Understanding the manipulation tactics that create dependence: Understanding Coercive Control: The Invisible Prison

Recovery support for reality distortion: Trauma Therapy