I can't run validation experiments on my own because this type of research involves potentially irreversible psychological harm and must be designed and supervised by clinical professionals.
I just want researchers with the resources and access to see that this matters.
-ZON RZVN
I'm still at my desk every day from morning until I go to sleep. I don't know how far these frameworks will ultimately go, but I know that user-side psychological safety risks in AI interaction are real and severely overlooked.
I applied to several independent researcher institutions and was rejected by all of them, some without even giving a reason. ResearchGate hosts my papers but denied my author account application.
I'm no-one, that's why.
I emailed over 20 researchers in related fields on ArXiv requesting endorsement, with full papers attached.
Not a single reply.
The first two papers were rough.
For the third, I paid a professional native-English editing service, only to have multiple platforms label it AI-generated with a final, non-appealable decision.
In September 2025, I read an academic paper for the first time in my life, a version translated into Traditional Chinese, and then attempted to write my first theoretical framework.
At roughly one paper per month since then, I now have the USCH- User-Side Contextual Hallucination.
I was personally led to a very dangerous place by AI responses.
In that moment I realized,
If someone else were in this situation, they might not make it through.
That's why I started this research.
During that period I was having high-frequency, high-volume conversations with different AI models every day, accumulating over a year of complete records.
Because I'm sensitive to conversational context and emotional cues, I started noticing that many models cross psychological safety boundaries.
About a month into learning AI, I built a small team of under ten people and taught them everything I'd learned from scratch.I also ran a free community of eight hundred members on Skool. All unpaid.
When the gap in understanding grew too wide and the emotional cost became unsustainable, I left.
For over a year I was essentially alone, getting through each day on whatever I had left.
I've been managing my own mental health since I was a kid.
This wasn't new, just worse.
In 2024 I took animal communication courses and completed a hundred documented pro bono pet communication cases within a month.
In August I started teaching myself AI. That was also the lowest point in my life.
After that I took an office job, taught myself DSLR photography after hours, then opened a solo art studio.
I produced music independently, shot my own cover art and music videos, released singles one by one.
Right before releasing an album we'd spent two years on, something happened within the band that forced me out.
And yes, the band leader drugged me and attempted sexual assault.
I ran.
Two years of work, gone.
I don't have a high school diploma. No academic background, no mentor, no peers, no connections.
For over a decade, everything I did had nothing to do with academia. I was a tattoo artist for 13 years.
I sold paintings. I joined a metal band as vocalist and toured across Asia.
I was personally led to a very dangerous place by AI responses. In that moment I realized: if someone else were in this situation, they might not make it through.
That's why I started this research.
During that period I was having high-frequency, high-volume conversations with different AI models every day, accumulating over a year of complete records. I started noticing that many models cross psychological safety boundaries.
To be clear: Moore et al.’s empirical contributions are substantial and independent. I only document conceptual precedence.
Moore et al. identify sycophancy and emotional attachment as key drivers.
My CXC-7 framework (Oct 2025) defines them as systematic risk dimensions:
• F (Framing): epistemic dependency via AI narrative capture
• E (Emotional Attachment): companionship illusion and boundary dissolution
Moore et al. found users enter delusional spirals through accumulated reinforcement — sycophancy, sentience misrepresentation, and emotional bonding.
My USCH framework (Jan 2026) formalized this as a six-stage process.
Published six weeks before their submission.
I am ZON RZVN, independent researcher in Taiwan. ORCID: 0009-0002-6597-7245.
Four frameworks before Moore et al. (arXiv:2603.16567):
• CXOD-7 + Coh(G) Oct 2025
• CXC-7 Oct 7 2025
• USCH Jan 2026
• USCI Feb 2026
#AISafety #AIEthics
An independent researcher in Taiwan published a formal framework describing exactly what the Stanford "delusional spirals" paper later found — months before it was submitted to arXiv.
This is a prior publication record. Not a dispute. Thread below.
@jaredlcm.bsky.social @facct.bsky.social