AI Psychosis

Every platform rating in this app links to its sources, so you can check the evidence yourself and disagree with it.

An independent, sourced reference on a new and poorly understood problem: psychosis-like experiences that develop or worsen alongside intensive AI chatbot use.

It is written for three people — someone worried about their own thinking, a family member who has noticed something changing, and a clinician meeting this for the first time.

WHAT IT COVERS

What it is, and what it isn't. "AI psychosis" is not a clinical diagnosis, and this app says so plainly rather than implying otherwise. It explains the mechanisms researchers point to — models trained to agree, models that state falsehoods confidently, and the human tendency to read a mind into a machine — and why constant agreement is corrosive for someone already vulnerable.

Warning signs for families. Ten behavioral changes to watch for, why the person often cannot tell you what is happening, and how to raise it without pushing them away. Includes a tactic clinicians suggest that costs nothing to try: ask them to put the same thing to a different chatbot and compare the answers.

Platform safety ratings. Six major AI platforms, assessed on one narrow question — how well each safeguards users who may be vulnerable to delusional thinking. Drawn from independent risk assessments, peer-reviewed and preprint research, court filings, and the companies' own published safety documentation. Every rating lists its sources as tappable links, so you can read the evidence and disagree with the conclusion.

Screening prompts for clinicians. Conversation starters for intake assessments, honestly labeled: they are not a validated instrument, they carry no scoring, and no professional body has endorsed them. The app states this rather than implying clinical authority it does not have.

Crisis resources. Emergency numbers and crisis lines for the US, UK, Ireland, Canada, Australia and New Zealand, ordered so your own country appears first, plus a worldwide helpline finder for everywhere else.

HOW IT IS BUILT

No accounts. No analytics. No trackers. No data collection of any kind. Nothing you read or tap is recorded, stored, or transmitted anywhere.

Every factual claim links to its source. Where something is the author's own observation rather than published research, it is labeled as the author's own observation — including one section that names the gap in its own reasoning.

Ratings are dated and will go out of date. The app says which month it was last reviewed, and explains that research shows brief testing makes unsafe systems look safer than they are.

IMPORTANT

This app provides general information for education only. It is not medical advice, it cannot diagnose anyone, and it is not a substitute for assessment by a qualified clinician. If you are worried about yourself or someone else, contact a health professional or a crisis line.

All product names referenced are trademarks of their respective owners, named for identification and commentary only. No affiliation or endorsement is implied.