GiveCare AI / policy taxonomy sketch

Care AI sits in the pressure field between existing regimes.

It is not clinical medical AI, not therapy, not only a companion chatbot, and not only a sensitive-data product. It inherits risk signals from all of them.

Care AI regulatory terrain A terrain map showing caregiving AI at the center of overlapping regulatory pressures from clinical medical AI, therapeutic AI, companion AI, health-data law, and high-risk AI governance. Higher care vulnerability More formal legal duties Clinical medical AI diagnosis, treatment, triage, clinical safety Therapeutic AI therapy boundary, crisis posture, false care Companion AI attachment, minors, simulated intimacy Data and high-risk AI sensitive data, impact assessment, audit Caregiving AI non-clinical support used by vulnerable people over time Missing Care AI standard The overlap is by risk concern, not by product category. A care assistant can be non-clinical and still trigger clinical, therapeutic, companion, data, and audit expectations.

Clinical similarity

Care AI may discuss symptoms, medication routines, or care decisions even when it does not diagnose or treat.

Therapeutic similarity

It may receive grief, guilt, suicidal ideation, loneliness, or emotional dependence without being a therapist.

Companion similarity

Persistent memory and warm tone can create attachment, identity ambiguity, or reliance under stress.

Governance similarity

Care context often includes sensitive health, family, financial, and vulnerability data that needs auditability.

The policy argument is not that caregiving AI already belongs neatly inside one existing law. The point is that it lives in the overlap of risks those laws are starting to name, which is why a national caregiving AI standard has to be relationship-aware, source-inspectable, and deployment-oriented.