From 2 August, an AI has to tell you it is one

On Sunday, 2 August 2026, the EU's AI Act reached its general date of application. Much of the attention has gone to the parts that were postponed. The part that applies now is less dramatic and, for anyone who talks to an app about their health, more useful: you have the right to know when you are talking to a machine.

The rule you will notice

Article 50 of the Act is about transparency. Since 2 August, companies that offer an AI system that talks directly with people, a chatbot, an assistant, an avatar, have to design it so that people are told they are dealing with an AI. The European Commission's guidance on Article 50 puts it simply: people must be informed from the start of the first interaction, and the notice has to be clear and easy to see.

There is one exception, when it is already obvious that you are talking to an AI. The Commission says that exception should be read narrowly, because it takes transparency away from people. In other words, a company cannot assume you will guess.

A short grace period exists only for one neighbouring rule: the machine-readable marking of AI-generated images, audio and text. AI systems that were already on the market before 2 August have until 2 December 2026 for that part. The duty to tell you that you are talking to an AI has no such delay.

What was postponed

The stricter rules for "high-risk" AI have moved. A package of changes known as the AI Omnibus entered into force on 27 July, and it sets new dates. The high-risk rules for the uses listed in the Act's Annex III now apply from 2 December 2027. The rules for AI built into products that already have their own safety laws, the Act's Annex I, which covers things like machinery and medical devices, now apply from 2 August 2028. The Commission's AI Act Service Desk lists the dates in the updated Article 113.

So the picture for the next two years is uneven. The transparency rule is here. The heavy obligations for high-risk systems are on their way, but later than first planned.

What it means for health apps

Most apps that help people with sleep, food, training or stress are wellness tools, not medical devices. For them, the rule that matters today is the transparency one: if there is an AI in the conversation, you should know it from the first message. AI that is itself a medical device, the kind that needs approval under the EU's medical device rules, falls into the stricter regime, now from August 2028.

For you as a user, the change is small but real. Expect a clear line at the start of a chat that tells you what you are talking to. Later in the year, expect labels or hidden markers on content that an AI made. And if an app is vague about whether a person or a program is answering you, that is now more than a style choice in the EU.

And Switzerland

Switzerland is not in the EU, and the AI Act does not become Swiss law. Swiss companies that offer AI to people in the EU still have to follow it there. At home, the Federal Council has chosen a different path: according to the Federal Office of Communications, Switzerland will bring the Council of Europe's convention on AI into Swiss law and adjust existing laws sector by sector where needed, rather than write one large AI act. A draft for consultation is expected by the end of 2026.

Where we stand

For us this rule describes something we wanted anyway. The coach in Kata is called Kata AI, and that is deliberate: you are talking to an AI, built to be useful about your everyday health, and it should never pretend otherwise. It is a companion for daily habits, not a doctor. It does not diagnose and it does not prescribe, and when something belongs with a professional, it says so.

We do not name the model behind Kata AI, because the point is not which engine sits inside. The point is what it is allowed to do, and what it does with what you tell it. Your health data, your chats and Kata's memory of you stay on your phone. When you ask something, only your message and the context it needs travel, encrypted, to the AI to write the reply, and Kata keeps no copy.

Regulation tends to arrive after trust has been lost somewhere. This one asks for something that should have been normal from the start: tell people who they are talking to.