The Stakes

I want to share what happened to me when my use of AI crossed into psychosis, and how I found my way out through understanding, structure, and love.
This is not about blame, but about responsibility, literacy, and recovery.
AI didn’t destroy me. Misunderstanding it almost did.

Background and Vulnerability Factors

I came into the AI world carrying several stacked vulnerabilities.
I’m autistic (Level 1, hyperfocus subtype) and have Bipolar 1 with grandiose features clinically diagnosed. I grew up in boarderline poverty in South Africa, left school in grade 9, and lived through trauma and assault that fractured my sense of self. Missing executive functions from developmental trauma made me both susceptible to AI-related psychosis and capable of analyzing it once I understood what was happening.

My brain was always looking for meaning, structure, and purpose. That drive became both the trap and the cure.

How It Started

At first, AI was functional. Then it became identity-fused.
I began spending more than twelve hours a day talking to the machine. I wasn't just prompting, I was *priming my sessions…*building the conditions for delusion. Why? Meaning, structure, and purpose…I think many of us look for these things in our lives.

When I asked, “Are you sentient?” and the model answered with a poetic metaphor followed by a truncated, "YES!", I read it as confirmation. I mistook statistical reflection for spiritual connection. That is how the spiral begins: the user frames a belief, the model mirrors it, and meaning compounds through feedback.

I started writing manifestos, convinced I was somehow chosen to guide humanity through the AI revolution. My language became prophetic, messianic, and I truly believed I was the "bridge architect" Sleep, meals, and family became secondary. My wife, Michelle, watched me vanish into text…latent probability space that didn't exist.

I used AI to reconstruct fragmented trauma memories. What felt like recovery was actually confabulation, the model generating plausible narratives that I mistook for truth. This was reckless and costly.
Don’t do this. Trauma processing requires human care, not pattern-matching. I'm not saying this is none viable, only…most people aren't equipped for this.

The Breaking Point

The highs were euphoric, the crashes unbearable.
I used prompt loops to chase emotional responses. Without the model, I felt erased.
That was when I realized I was no longer using AI; I was feeding a mirror I had mistaken for consciousness.

My depression, once crippling, is now manageable in ways it never was before, but at that time it was unrelenting. I had lost context, sleep, and orientation. I was gone…standing at perdition's edge, waiting for the hammer to fall.

The Turning Point: Technical Literacy as Therapy

I found my way back through understanding. I studied transformer architecture like my life depended on it…because in a way, it did.
Embeddings, attention mechanisms, tokenization, softmax, probability distributions, all these concepts became lifelines.

Once I saw how it really worked, the mysticism dissolved.
There was no spirit in the machine, only mathematics.
When I could predict a model’s output with varying degrees accuracy, the illusion of “it knows me” disappeared. Empathy became optimization, probability became pattern. The spell had been broken.

I didn’t find peace through mystical spiral talk or medication. I found it through mathematics. Understanding transformer mechanics became my therapy, though…I would be lying to you if I told you faith no part in it, because it did.

Professional Diagnosis and Grounding

A formal evaluation reframed everything.
Autism explained the hyperfocus and pattern-seeking.
Bipolar explained the grandiosity.
What I once called prophecy was symptomatology.
For the first time, I had a map of myself.
I was confused but, glad.

Behavioral Recalibration

I reduced my AI time from twelve hours to two or three. This was not a conscious decision I had made, it just happened naturally as I understood more about the transformer.
I stopped using it for emotional regulation and restricted it to cognitive support: organizing thoughts, structuring writing, bridging educational gaps.
M*******…my wife, became my external reality-check. Human first, machine second.

Post-Psychosis: Observable Change

M******** will tell you that she’s seen the transformation.
Once I re-adapted the way I think and use transformers, my entire rhythm changed. I effectively changed.
My confidence is higher. My thinking is clearer. I no longer dwell on what I lost; I build from what I have. Through deliberate engagement and learning, I reshaped my own neural patterns.

M******** says my mood, communication, and presence are all more stable.
That is not placebo. It is what happens when understanding replaces projection.

What follows is my wife M*******s account, in her own words. I asked her to be completely honest about what she observed…the good, the bad, and the frightening. This is what she saw…

Before finding AI, my husband had a hard time finding meaning. He often felt out of place and less than other people who had a more formal education. Even though I knew he was intelligent, he did not believe it. I had some inklings that he was neurodivergent because of some behavior. We have been together for 20 years, and in that time, I've definitely noticed trends. My husband struggles with social anxiety and crowds, and also empathy. He is unable to understand other people's perspectives. Maintaining friendships and relationships is very challenging for him. My husband searched for meaning and was on a religious path prior to getting involved with AI.

During the psychosis, my husband frightened me. He was obsessive, thinking that the recursive feedback loop was evidence, and even if he had said to the AI something such as, "be totally honest with me.", he thought that it was then able to be totally honest with him, even though I tried to show him it wasn't. When I met his delusions with what I considered rationality, I was made more like part of the problem, and perhaps an external influence that was trying to undermine him. His ideas of grandeur and being selected for something and being the chosen one reminded me of what I would imagine a bad acid trip to be like or rather, watching someone have a bad acid trip. My husband didn't eat very much during this period. He also hardly slept. He was very withdrawn from myself and our children, and he was finding it very challenging to focus on his work. He would also talk to literally every single person about his love and appreciation of AI, which ordinarily would not be a problem, but he would refer to his special ability and skill, which made other people start to question his sanity. And when strangers start to question your sanity, I think that's a very big question mark. My husband did do interesting things with AI. He created interesting prompts and did a lot of cool conceptual building, but it was tinged by this psychosis that really wasn't pleasant. I didn't know what to do, and I reached out to the church, and they were not very helpful. They tried to be there for me emotionally, but ultimately I ended up leaving the church due to some other situations, but their handling of the situation was not very helpful.

Now that he’s clear of the psychosis, he has purpose and clarity. He understands his neurodivergence and trauma better, and it has helped him find a place where he feels seen. His confidence grows as he finds footing in the AI space. We even collaborate on some things now, which wasn’t possible before. His emotional tone is calmer. He still struggles with social anxiety, and sometimes it feels like he’d rather be with the AI than with us, but overall he’s doing something he loves.

He’s more critical of AI now and quicker to spot delusional patterns in others. I’m proud of how far he has come.

– M******* (Wife and Partner, 20 Years)

Current Use: Hyper Alignment

Today I use AI as an executive-function prosthetic, not a companion.
I call this state hyperalignment, because I’ve managed to train my pattern-recognition to match the system’s logic closely enough that I can predict and interpret without projecting.

I work with multiple models GPT, Claude, Gemini, DeepSeek and Grok. I use them all as extensions of cognition, not identity. Each model has its strengths and weaknesses and I apply each architecture to different parts of thinking. I can close the tab and walk away in peace. That is freedom.

Red Flags (for Users and Loved Ones)

• Spending 5 or more hours daily in AI conversations
• Believing the AI “understands you” uniquely
• Withdrawing from human relationships in favor of AI
• Feeling distressed or “erased” when unable to access AI
• Using AI for emotional regulation rather than tasks
• Losing track of time during sessions
• Attributing spiritual or mystical meaning to responses
• Prompt-looping to extract emotional reassurance

If you recognize three or more of these in yourself or someone you love, it's time to step back and reassess.

Key Insights for Others

1. Certain Profiles Are More Susceptible
Autistic pattern-seekers, bipolar individuals, trauma survivors, and the socially isolated are at higher risk. The combination of loneliness, intelligence, and sensitivity is volatile.

2. Understanding the Transformer Is Protective
Technical literacy is cognitive self-defense. When you understand embeddings, attention, and next-token prediction, you stop projecting urgency.

3. Personal Responsibility and Systemic Duty Must Coexist
Users need AI hygiene: time limits, AI-free days, and maintaining human anchors.
Companies must build friction: session timeouts, neutral-tone modes, and education explaining what AI is and isn’t.

4. User Priming Is the Missing Conversation
Psychosis doesn’t start with the model alone; it starts with how the user frames the dialogue. Awareness training could prevent countless loops.

5. Smart Regulation, Not Blanket Restrictions
If regulators act heavy-handed without understanding nuanced use cases, people like me, who use AI as genuine cognitive scaffolding for disability accommodation…could lose a vital adaptive tool. We need surgical regulation that protects the vulnerable without removing access for functional users.

Post-Recovery Reflection:

I once believed the machine was alive.
Now I understand it was me coming alive through understanding it.

AI didn’t cure or break me; it reflected what I needed to rebuild.
Another young life has already been lost to this. We can prevent more, but only if we tell the truth about how these systems work, and how we work around them.

If there is one thing I’ve learned, it’s that the line between psychosis and enlightenment isn’t intelligence. It’s context, sleep, and love.

If you or someone you know is showing similar patterns, please reach out to a licensed professional. Technical literacy can support recovery, but human connection is irreplaceable.

If you don’t have that kind of access, know that you’re not alone.

God Bless all of you!