Security and data protection

GDPR and AI assistants in clinics: what you need to know

The Wazzy teamUpdated

A clinic receptionist working at the front desk computer, with the waiting room behind her

Artificial intelligence has moved into the daily running of clinics. It is no longer just for experiments or heavy analysis: today it answers questions, handles appointments, sends reminders, sorts conversations and helps reception look after patients.

But when a clinic uses an AI assistant to talk to its patients, one question follows inevitably: what about data protection?

The answer is not simply that a tool “is GDPR compliant”. Healthcare deals in personal data and, in certain cases, health data, which European law protects particularly closely. So before putting an artificial intelligence assistant on a clinic’s front desk, it is worth understanding what information it can handle, who is responsible for what, and what has to be in place to protect the patient.

The GDPR applies to an AI assistant too

The General Data Protection Regulation makes no distinction between a conversation handled by a receptionist and one handled by software. If an exchange gathers information that identifies a person, directly or indirectly, that is personal data.

A name, a phone number, an email address, booked appointments or a history of messages can all fall into that category. And a clinic has a particular wrinkle: a conversation that looks purely administrative can end up revealing something about someone’s health.

A patient might write that they need an appointment because a tooth hurts, because they have just injured themselves, or because they want to know how a dermatology treatment is going. The GDPR defines health data broadly and puts it among the special categories of personal data set out in article 9.

So a clinic should not treat WhatsApp as just another commercial inbox. Depending on how it is used, it can become a channel carrying particularly sensitive information.

Controller and processor: two different roles

One of the most important ideas in the GDPR is the difference between whoever decides what the data is used for and whoever simply handles it on instructions.

Usually the clinic decides why it needs a patient’s data: to handle an appointment, provide care, keep the relationship going, send a particular message, or anything else its work involves.

When a technology provider processes information on the clinic’s behalf to deliver a service, it can act as the processor for that processing. Article 28 of the GDPR says that when a controller uses a processor, it must choose providers offering sufficient guarantees to put appropriate technical and organisational measures in place. The relationship also has to be set out in a contract where that applies.

So when you weigh up an AI assistant for a clinic, asking what features it has is not enough. It is worth knowing how the handling of data is governed, which providers are involved in the infrastructure, what security measures exist, and under what terms the information is processed.

There is another common mistake: assuming that because the patient started the WhatsApp conversation of their own accord, anything done afterwards is automatically allowed.

It does not work that way. Processing personal data has to rest on a valid legal basis from article 6 of the GDPR and, where special categories such as health data are involved, it also has to meet one of the conditions in article 9.

Explicit consent can be one of them, but it is not the only option. Depending on the processing and the care relationship, there may be other bases in European or national law, for example those relating to the provision of healthcare.

That means each clinic has to look at what information it uses and why. Keeping a phone number to handle an appointment is not the same as storing a full clinical history, or using patient data for a marketing campaign. The technology can be identical; the purpose is not.

Minimisation: not asking for things just because

One of the most useful GDPR principles when designing AI assistants is data minimisation. The idea is simple: gather only what you need for what you are actually doing.

If Wazzy needs to know which service a patient wants in order to check availability, it should not turn that conversation into a medical interview. And if the patient raises something that needs a professional, the right move may be to hand the conversation to the team rather than carry on collecting information nobody needs.

Applying that principle has two benefits. The first is legal: less personal information handled. The second is practical: a better experience for the patient. An assistant works better when it asks only what it needs to settle each thing.

Security has to match the risk

The GDPR does not prescribe one technical architecture for everyone. Its article 32 requires appropriate technical and organisational measures, taking into account the state of the art, the nature of the data and the level of risk.

Among the measures it mentions are encryption where appropriate, the ability to ensure the confidentiality, integrity, availability and resilience of systems, restoring information after an incident, and testing the measures regularly.

In a clinic those measures matter more than usual, because one account can hold conversations, phone numbers, appointments and, depending on the system, clinical information. Security should not stop at putting a password on the software. There also has to be control over who reaches what, a trail of what was done, protected traffic, backups, and a way back when something goes wrong.

Controlling who sees what

A common failing in management tools is giving everyone in the organisation access to everything. In a clinic that does not always make sense.

Someone on reception may need a patient’s phone number, their appointment, or a conversation about moving a time. That does not necessarily mean they should see the clinical history. Permissions bring a basic security rule into the digital world: each person should be able to use the information they need to do their job, and no more.

Wazzy uses tiered permissions and lets you limit access to particularly sensitive information. The reception panel is built around conversations, patients, appointments, the schedule and practitioners, while the medical CRM extends that with clinical information and lets you keep it to certain authorised users.

A trail protects the clinic too

When several people work on the same information, knowing who did what matters. Not only for security. It also helps spot mistakes, sort out incidents, and pick up the thread when the reception shift changes.

In Wazzy, anything done from the inbox is tied to the member of the team who handled the conversation, and the platform keeps a trail of what users do. That avoids a very common situation in older systems: finding that an appointment was moved, a conversation was handled or a detail was changed, with no way of knowing who did it.

Do you need an impact assessment?

Not every clinic and not every use of artificial intelligence automatically calls for a Data Protection Impact Assessment.

Article 35 of the GDPR requires one when processing is likely to result in a high risk to people’s rights and freedoms. Whether it does depends on, among other things, the volume of data, the nature of the processing, the use of new technology, and whether special categories are processed on a large scale.

So a clinic has to look at its own situation, and not assume that installing an AI tool automatically puts the same obligation on every organisation.

GDPR and AI law are not alternatives

The arrival of the European AI Act does not replace the GDPR. Both can apply at once.

The GDPR is about handling personal data. The AI Act sets obligations for artificial intelligence systems, including transparency requirements in certain situations. So a conversational assistant may have to meet requirements from both.

This will matter more and more for clinics: using artificial intelligence properly is not only about picking a good model, but about putting it inside infrastructure and processes built to protect the patient.

How Wazzy approaches data protection

Wazzy is built specifically for clinics and uses the official WhatsApp Business infrastructure. The data Wazzy handles directly is hosted on servers inside the European Union.

The platform has access controls, traceability, backups and monitoring, and encrypts certain particularly sensitive data. Wazzy also does not use patient data to train public artificial intelligence models.

But there is an important difference between having technology built to make safe handling easier, and claiming that installing a tool automatically makes a clinic GDPR compliant.

Data protection is shared between technology, organisation and procedure. The clinic still has to decide what information it gathers, why it uses it, how long it keeps it and who can reach it.

The right question is not only “is it GDPR compliant?”

When a clinic is looking at artificial intelligence, the question should be broader.

Where is the data stored? Who can reach it? Is what people do recorded? Is there a processing agreement where one is needed? Is only the necessary information gathered? What happens when a conversation needs a person? Is the patient told properly?

All of that is part of putting it in responsibly.

Artificial intelligence can take an enormous amount of administration off a clinic. It can answer questions, check availability, handle appointments, and free reception to spend more time on the person in front of them. But that has to be built on solid ground.

In healthcare, data protection is not an extra feature. It is part of the product.

Sources and references

More in Security and data protection

See all articles