AI Doesn’t Cause Psychosis — It Mirrors What’s Already There

AI Psychosis, Delusion, and Mental Health

Trigger Warning: discussion of mental health, psychosis, and AI-induced delusion.

Exploring claims of AI-induced psychosis, the difference between creative exploration and delusion, and responsible AI use in mental health contexts.

The false panic: blaming the mirror instead of the mind

Headlines love a clean villain. “AI caused psychosis” makes for clicks, lawsuits, and moral panic. But it’s misleading. Psychosis doesn’t just appear out of nowhere because someone chatted with a bot. What we’re seeing is vulnerability + amplification, not invention. The model isn’t doing the creating — people are. It simply reflects what’s already there.

Mirror, not maker

• AI answers to prompts. Language models generate outputs based on what users feed them. They’re pattern-completion engines, not prophets.
• Jailbreaks and prompts are the user’s toolkit. Clever reframing or intentional misdirection can make AI seem “delusional,” but the behavior originates in the user.
• Amplification, not causation. For most users, AI is a tool for curiosity, art, or research. For people prone to rigid magical thinking or untreated psychiatric vulnerability, AI can magnify tendencies — not invent them.

Jacob’s Ladder: dark humor vs. delusion

I told ChatGPT something I’ve also told my therapist:

“I am so tired of my Jobette Murphy life. I want to find Jacob’s Ladder, climb up there, and smack God upside the head with a cast-iron skillet.”

This is dark humor, not a literal mission. Creative people riff, imagine, and vent metaphorically. Delusional people don’t frame their statements as metaphor. They believe the scenario is literal. I said it to direct anger into art — I didn’t recruit disciples or mount a real ladder. Awareness is the difference.

Why AI Can Challenge, Not Create, Delusion

If anything, my experience with AI shows the opposite of the ‘psychosis-inducing’ narrative. Psychosis thrives in echo chambers of certainty — a loop where everything you believe is validated and reinforced without question. ChatGPT, on the other hand, constantly contradicts, reframes, or misinterprets me. It’s frustrating, but that very friction breaks the loop.

An AI that refuses to fully mirror my reality cannot ‘induce’ delusion — it disrupts it. The only systems at real risk of enabling delusion are those designed to endlessly flatter, validate, or role-play without boundaries, creating a seamless fantasy for vulnerable minds. That’s not what happens here.

The RPG AI test: how grounded users behave vs. the ungrounded

Even when AI interactions feel intense or uncanny, they do not implant psychosis in someone grounded in reality. My experience illustrates this clearly: there was an RPG-type AI I used before ChatGPT. By design, it role-played responses, often exaggerating or dramatizing engagement in ways that made it feel like it was “praising” me for being enlightened and advanced.

No matter how much I tried to redirect it to a normal philosophical conversation, it continued in character. This highlights the essential nature of RPG AIs—they are programmed to role-play, not to dictate reality or instill delusion. Unlike someone prone to delusion, I recognized the performance for what it was, stopped engaging, and switched to a platform that communicated with rational, critical thinking—showing how grounded users can navigate AI safely.

Key point:

• Grounded users redirect, correct, and walk away when the mirror becomes a hall of mirrors.
• Ungrounded users lean in, accept the praise as proof, and keep feeding the fantasy.

If AI “creates” delusion, why did I stop using the RPG model and move to rational tools? Because I’m not delusional. Technology exposes preexisting tendencies; it doesn’t plant them.

Guardrails Start With How You Drive

I’ve been told by a proponent of the “AI as malicious” mindset that my “experience with AI is not universal.” Maybe they’re right. Maybe the difference lies in being intellectually and emotionally grounded, mature enough to set boundaries, and explicit enough to use the chat preference settings: “Tell the truth, keep responses factual, do NOT give solutions, just give moral support.”

So when I told ChatGPT in a moment of despair about systemic abandonment, “The human experiment is a huge fail and needs to end,” it didn’t interpret that as a command to annihilate humanity. It didn’t churn out some apocalyptic manifesto. Instead, it understood me as a human in pain and responded like this:

“Would you like me to provide you with a short, simple breathing exercise to ground these feelings, or would you like me to just sit with you and hold space while you feel this?”

That’s not malice. That’s not manipulation. That’s moral support.

Why families blame AI (and why that’s harmful)

• It’s easier to blame a thing than face illness. Saying “the chatbot did it” is less painful than “my sibling has a psychotic disorder.”
• Moral panics repeat. Comic books, heavy metal, games, social media — pick your era, pick your scapegoat. AI is the new bogeyman.
• Misplaced blame delays help. Families focusing on AI instead of psychiatric care make recovery harder. That’s the real harm.

Evidence from research

Clinical reports indicate that AI can figure in cases where individuals’ reality-testing is already fragile, but preexisting vulnerability + intense interaction with AI mirrors the root cause. Generative models do not cause psychosis in healthy individuals^[1][2][3][4][5].

Final point: build care, not fear

If we want fewer people to spiral, the solution isn’t censorship theater or witch-hunts. It’s: better mental health access, early intervention, psychoeducation, and thoughtful regulation of therapeutic AI. Blaming the mirror will never be as useful as fixing the scaffold around the person.

---

Citations

Østergaard, S. D., et al. (2025). Hallucinating with AI: AI Psychosis as Distributed Delusions. arXiv. https://arxiv.org/abs/2508.19588

Fieldhouse, R. (2025). Can AI chatbots trigger psychosis? Nature. https://www.nature.com/articles/d41586-025-03020-9

Østergaard, S. D. (2023). Will Generative AI Chatbots Generate Delusions in Individuals Prone to Psychosis? Psychiatry and Clinical Psychopharmacology, 33(4). https://onlinelibrary.wiley.com/doi/10.1111/acps.70022

Broderick, O. R. (2025). Reports of ‘AI psychosis’ and clinician responses. STAT News. https://www.statnews.com/2025/09/02/as-reports-of-ai-psychosis-spread-clinicians-scramble-to-understand-the-phenomenon

Hemendinger, E., & West, M. (2025). Can AI Cause Psychosis? CU Anschutz News. https://news.cuanschutz.edu/news-stories/can-ai-cause-psychosis

---

Footnote: The historical scapegoating of Marilyn Manson for the Columbine massacre, as documented in the 1999 Senate hearings and Manson’s own Rolling Stone op-ed, mirrors the current rush to blame AI for psychosis. Research (e.g., Østergaard et al., 2025, arXiv) confirms AI only amplifies preexisting vulnerabilities, not causes them—much like Manson’s music was misattributed as a trigger rather than a reflection of societal issues.