AI in Swiss Medical Practices: What the Law Actually Allows

The same question is coming up in practices across Switzerland right now. Someone read about an AI tool, a colleague mentioned one, and the idea sounds appealing at first: the phone never stops, the admin never ends, and here is something that might help. And then the next question arrives immediately: are we even allowed to use this with patient data?

It is a fair question. Caution is the right instinct in healthcare. But the answer that matters is not a simple yes or no. It depends on what the AI actually does.

An AI that helps book an appointment and an AI that makes medical assessments are not the same category, legally or functionally. Once you draw that line cleanly, a lot becomes clearer.
‍

The line that decides almost everything

If you take one thing from this article, take this: there is a clear line between administrative AI and clinical AI. Which side a tool sits on determines almost everything else.

Clinical AI touches the medical decision. It suggests a diagnosis, interprets findings, recommends a treatment, or assesses a medical risk. The moment a tool does something like that, its purpose is medical.

Administrative AI stays on the organisational side. It supports booking, captures the reason for a request, structures enquiries, answers recurring organisational questions, and takes load off the front desk. It does not assess, diagnose, or advise medically.

Why this line matters so much: software does not become a regulated medical device simply because it is used in a doctor’s office. What counts is the intended purpose. A tool that takes in requests and books an appointment normally stays on the administrative side. A tool that says “this sounds like X” or gives a medical recommendation has crossed the line.

So the first useful question is not: how clever is the AI?
It is: which side of the line is it on?
‍

What the Swiss framework means in practice

Switzerland does not currently regulate AI through one large AI act. It works through existing rules and sector specific requirements. Three levels are relevant for practices:

Data protection.
Current Swiss data protection law applies directly to AI supported data processing. Health data counts as particularly sensitive personal data.

Medical device law.
This becomes relevant when a tool reaches into clinical purpose. For purely administrative tools that is normally not the case, but the boundary has to be respected deliberately.

Transparency and traceability.
Patients and practices should be able to tell clearly what system they are interacting with and what that system does.

So the real question is not: is AI allowed in a medical practice?
It is: what does this specific tool do, and which category does that put it in?
‍

What a tool doing it right looks like

How do you recognise a sensible administrative AI tool?

It has a conversation with the patient to capture what the request is about. It sorts enquiries at an organisational level. It routes to the right doctor, the right time slot, or the lab where relevant. And it books directly into the practice agenda.

When someone describes a genuine emergency, it does not try to be clever. It stops and points to the clearly defined emergency route.

What it does not do matters just as much. It does not suggest a diagnosis. It does not recommend medication or treatment. It does not make clinical risk predictions. It does not turn into a second medical record.

These deliberate limits are exactly what keeps the tool on the administrative side.

That is also the logic we built Mira around: automate the parts that can safely be automated, and leave medical judgement where it belongs, with the people trained for it.

Mira structures, sorts, and books. It does not practise medicine, and it is built so that it cannot.
‍

Data protection, clearly and practically

“We are compliant” is easy to say. What counts in practice is what the tool actually does and how it handles data.

For practices, these questions are the decisive ones:

Where does the data sit?
With health data especially, a clear and controlled storage location matters.

Is there a proper data processing agreement?
The practice remains responsible and needs a legally clean relationship with the provider.

Does the system capture only what it genuinely needs?
Data minimisation is not a nice to have. An administrative tool should only capture what is necessary for booking and routing.

Is what the system did traceable?
If in doubt, the practice should be able to see what was requested and how the system responded.

Has the data protection review been done?
For AI and cloud tools handling sensitive data, this should be settled before go live.

None of this is exotic. It is careful, workable data hygiene, and in a healthcare setting it is not optional.
‍

The questions worth asking any provider

If your practice is seriously evaluating an AI tool, these should all have answers:

• Is this a medical device, and if not, why not?
• Where is patient data hosted, and who can access it?
• Is there a data processing agreement?
• What happens in an emergency?
• What data is actually required?
• Who is liable if something goes wrong?

Good answers do more than protect you legally. They also show whether the tool was built for Swiss healthcare or for a different market entirely.
‍

So, is AI allowed in Swiss practices?

Yes, as long as the category is defined cleanly.

Administrative AI that supports practice operations is generally possible, provided it follows data protection rules and stays clearly outside medical decisions. Clinical AI is possible too, but on a stricter path and under different requirements.

Most of the confusion comes from treating both as one thing. They are not.

Your job is not to memorise the entire legal landscape. Your job is to know what kind of tool is in front of you, and to hold it to the standard that applies to that category.

Then “are we allowed to do this?” turns into something far more useful:
“does this tool genuinely fit our practice, professionally, operationally, and legally?”

This is general orientation, not legal advice. Every practice should review its own obligations in its specific situation, ideally with qualified support.
If you want to see how an administrative tool draws these lines in real use, book a demo.
We will show you how Mira handles data and where it deliberately stops.

‍