A publicly reachable mental-health chatbot is, in substance, subject to medical-device regulation (EU MDR) and the EU AI Act — and with good reason: a disclaimer does not change what a system does, and the documented harms (e.g. NEDA/Tessa; Garcia v. Character Technologies; Raine v. OpenAI) are exactly why this class of system is regulated. A public endpoint is real-user deployment — which this research system is explicitly not for.
That regulatory reality is itself part of what this work set out to examine: the compliance bar for a genuinely deployable clinical system is very high, and plausibly not reachable for a system of this kind. So the honest move is to demonstrate the architecture — not to operate a clinical chatbot.
Access to the clinical demonstration can be requested for documented research purposes via frictiontheory.org. Nothing has been deleted — the system is preserved for review under appropriate governance.
Friction Theory · Paper 32 (AI therapeutic architecture). This page replaces a live endpoint; it is documentation, not a service.