The line that decides everything
Every one of the six regimes keys off the same thing: intended medical purpose. The presence of AI never triggers regulation. The claimed purpose does. That gives one line, and it runs through this entire reference.
The agent routes by what the patient says they want. Service interest, intake details, booking, follow-up. Administrative and commercial, never clinical.
The agent answers a symptom question in general educational terms. Safe only while it stays one-to-many and never interprets the individual.
The agent acts on the patient's data to reach a clinical conclusion: interprets symptoms, assigns urgency, suggests a diagnosis, or says no doctor is needed.
The safe zone and the forbidden zone are identical in all five primary markets plus the UK, because they all trace back to the same IMDRF definition. There is no laxer jurisdiction to hide in. Whatever clears the strictest market, Australia, clears all of them. The risk was never the architecture. The risk is the language and the stated purpose.
Route by what the patient asks for. Never route by what the agent concludes about their symptoms.
Keep the agent on the first side of that line, scrub the word triage from every patient-facing and contractual surface, and the device line stays behind us everywhere. Two outputs only ever exist on the patient path: schedule, or connect to a human. Do nothing and no need to see a doctor are the most dangerous outputs and are never permitted.