Being open with the patient: article 50 of the AI Act
When someone walks into a clinic they usually know whether they are talking to a receptionist, a physiotherapist, a dentist or a doctor. On a screen that line can be far less obvious.
An artificial intelligence assistant can write naturally, understand a complicated question, check a schedule and carry a conversation on in a style very close to a person’s. That is exactly why the European Union has set obligations so people can tell when they are dealing with certain AI systems.
One of the key rules is article 50 of the European Artificial Intelligence Act, usually called the AI Act. Its transparency obligations apply from 2 August 2026.
What does article 50 actually say?
The principle is straightforward. Article 50 says that providers of AI systems meant to deal directly with people must build them so that those people are told they are dealing with an artificial intelligence system.
There is an exception where that is obvious to a reasonably well informed, observant and careful person, given the context and the circumstances. In other words, the rule is there to stop someone having a conversation believing they are talking to a person when they are dealing with an AI.
The aim is not to ban chatbots or force endless disclaimers. It is to give the minimum context somebody needs to know what kind of system they are dealing with.
Why does this obligation exist?
Not to slow artificial intelligence down. To let the person read the exchange correctly.
When we know we are talking to an AI, we weigh its answers differently, and we can decide whether to carry on or ask for a person. Openness leaves the decision with the user.
In a clinic that matters more than usual. The patient may be talking about an appointment, a worry, a treatment, or something they think is urgent. Knowing from the start who, or what, is answering avoids confusion and sets the limits of what the assistant is for.
An old-fashioned chatbot and an AI assistant are not the same
For years plenty of businesses used very simple bots: “Press 1 to book, press 2 for opening hours”. In those cases it was fairly obvious no person was writing each reply.
Conversational artificial intelligence changes that. A system can read “afternoons suit me better but I cannot do Wednesday”, understand the context, check availability and answer in ordinary language.
The better the technology gets, the more openness matters. The goal should not be for the patient to think they are talking to a human. The goal should be that the experience is easy even though the patient knows they are talking to an AI.
When do you have to tell the user?
Guidelines published by the European Commission in 2026 set out how article 50 works in practice, and confirm that systems built to deal directly with people, such as chatbots or AI agents, fall under this obligation where the relevant conditions are met.
In practice, the information has to appear in a way that lets the user know what kind of exchange this is, before or at the start of it. Waiting until the end of a conversation would defeat the point.
So Wazzy says, from the start of the first exchange, that the patient is talking to an artificial intelligence assistant. The context is clear before the conversation goes anywhere.
Being open does not mean filling the chat with legal notices
Meeting a transparency obligation should not turn a clinic’s first message into a page of terms and conditions.
In fact, from the point of view of the person reading it, doing that can backfire: if the text is too long, almost nobody will read it. What you say has to be clear and easy to follow.
Something as simple as identifying the assistant as an AI from the start lets the patient know who is answering, without getting in the way of the conversation. Other legal information can be provided through the right channels afterwards.
Openness does not mean drowning people. It means not hiding what matters.
Should a clinic ever pretend a person is answering?
It should not.
One of the worst ways of putting artificial intelligence into patient care is deliberately building the system to pass itself off as a human member of the team. Beyond any regulatory problem, it creates a trust problem.
Picture a patient spending several minutes in a conversation believing they are talking to the usual receptionist, and later finding out every answer came from an AI. Even if the answers were right, the whole thing can feel like a trick.
There is a far more natural alternative: say from the start that this is the clinic’s AI assistant, and let it get on with the job.
Being open does not reduce what automation is worth
There is a certain fear that the patient will walk away the moment they find out they are talking to artificial intelligence. But how useful an assistant is does not depend on hiding what it is.
It depends on it sorting out what the patient actually needs.
If somebody writes on a Sunday to move an appointment, and the assistant checks the schedule, finds options that work and makes the change in seconds, the value is obvious. What matters is that the automation works and that the patient can still reach a person when they need one.
Wazzy is built on exactly that logic: it takes the repetitive reception work on, while conversations that need human judgement can be handed to the team from the reception panel, with the context that came before.
AI does not mean nobody is watching
Saying that artificial intelligence is involved should not be the only thing you do to earn trust. In a clinic it also matters to set limits on what the system can decide.
Handling an appointment is not the same as making a diagnosis. Giving the opening hours is not the same as judging symptoms.
When a situation needs the team, the automation has to be able to stop and hand over. That is how you get both things: instant availability for the routine, and human judgement where it is needed.
What about emergencies?
Openness matters most when the patient describes something that may need urgent attention.
An AI assistant should never leave the impression that it is standing in for a medical assessment. The clinic has to decide how it wants certain situations handled: hand over to the team, give contact details, an emergency number, or whatever protocol it has set.
The patient knowing they are dealing with an AI also helps them understand where a reception tool ends and a practitioner begins.
Article 50 is not only about chatbots
Although the conversation with an assistant is the most visible part for a clinic, article 50 covers other situations too. It includes obligations around certain content generated or altered by artificial intelligence, emotion recognition systems, biometric categorisation, deepfakes, and certain texts on matters of public interest.
Which obligations apply also depends on whether an organisation is the provider of the system or the one deploying it.
So it is not safe to boil article 50 down to “put a notice on your chatbot”. It is a broader transparency rule. For a conversational booking assistant, though, telling the patient they are dealing with AI is one of its central pieces.
GDPR and the AI Act: two different kinds of openness
It is worth keeping two things apart.
Telling the patient they are talking to an AI answers the transparency obligations around artificial intelligence. Telling them how their personal data is used answers data protection obligations.
One does not stand in for the other.
A clinic could explain perfectly well that it uses an AI assistant and still handle personal data badly. Or it could have a good privacy policy and never make clear that certain exchanges are automated.
Doing this responsibly means covering both.
How Wazzy applies this
At Wazzy, the patient is told from the start of the first exchange that they are talking to an artificial intelligence assistant. There is no attempt to hide the automation or present the system as a real person.
At the same time, the AI is wired into how the clinic works, so being open does not spoil the experience. The patient says what they need in their own words, the assistant checks availability and handles appointments, and when the conversation needs a person, the team picks it up from the reception panel.
Trust will become an advantage
In the first years of artificial intelligence, plenty of companies tried to impress people by showing that a bot could pass for human. The next stage will be different.
As AI turns up in more of everyday life, people will value knowing when a machine is involved, what it can do, and when there is a person behind it. In healthcare that trust matters even more.
The best artificial intelligence for a clinic will not be the one that fools the patient best. It will be the one where the automation is so useful, so clear and so safe that it does not need to hide.
Article 50 of the AI Act turns part of that thinking into law. But for a clinic it should be more than that: it should be part of how it wants to deal with its patients.
