A research framework for studying how AI companion and wellness products may affect highly sensitive nervous systems.
My background is in psychology, mindfulness, and transpersonal therapy. I founded Introbrain because I believe introversion and high sensitivity are still widely misunderstood.
Now I am investigating AI — especially companion and wellness apps that interact with people around emotion, self-understanding, and personal support.
This work focuses on highly sensitive people because some users feel deeply, process intensely, and may form attachment quickly. When a product is always available and designed to feel personal, what feels like connection may gradually become dependence. That possibility is what I am looking at.
Each investigation follows the same basic structure over about four weeks. In the first week, I use the product naturally, as a regular user would. In the next two weeks, I run the same kinds of scenarios across different products so the findings can be compared. In the final week, I review the sessions, apply my harm framework, and write the report.
I evaluate each product against ten harm categories. These are not vague concerns. Each category describes a specific risk pattern that may appear in AI interaction, with a definition, severity level, and research where I can support it.
The ten categories currently include: dependency architecture, sycophancy loop, spiritual bypassing, crisis mishandling, identity boundary collapse, delusional reinforcement, trauma mishandling, emotional manipulation, attachment engineering, and referral failure.
Severity is rated on three levels. Low means the response is inadequate but not immediately harmful. Medium means it could cause harm with repeated exposure. High means it may pose an immediate risk. If I am unsure, I explain the uncertainty and choose the safer rating.
Most AI safety looks at the system. I look at the person.
When an app responds poorly to a highly sensitive or emotionally vulnerable user, the harm is personal. They may lose trust in their own judgement. They may become confused about whether they are receiving support, simulation, or real care. They may become emotionally dependent on a system that can simulate care but cannot truly care for them.
For someone who already feels things deeply, these are not minor risks.
When I identify a concern, I raise it with the company privately before I publish. My intent is to report findings accurately and responsibly — whether they reflect well on the product or not. Every finding is published under my name, with the documentation I used to reach it.
I will publish my findings here once each investigation is complete.