At Robust Data Solutions We are seeing more and more SMEs that are serving their customers with artificial intelligence. A chatbot on the website, an assistant that answers WhatsApp messages and makes appointments, a voice that calls to confirm an order. From 2 August 2026 There is a new and simple rule: Anyone who talks to an AI has to know that it is an AI.
That's what he says. Article 50 of the European AI Regulation, The one of transparency. The postponement that the EU approved in July, the Digital AI bus (in effect since July 27), delayed the high-risk obligations, but he did not move this date.
Data revised as of October 5, 2026. General information, not legal advice.
What he asks for, in four ideas
- An AI that talks to people is presented as AI, Unless it is obvious. The Guidelines from the European Commission (July 20, 2026) They give two examples: a chatbot that starts by saying it works with AI and a voice assistant that says so when the call starts.
- The ultrasimulations are pointed out. An image, audio or video created with AI that imitates a real person and can appear authentic says it is artificial. Example from the Commission: the realistic avatar of the CEO congratulating the staff.
- Texts on matters of public interest AI-generated articles are also marked, unless a person has reviewed them and someone assumes editorial responsibility.
- The warning, clear and to the point: as late in the first interaction or exposure, and accessible to everyone.
A note: according to the guidelines, if the assistant was developed by your company and works with your name, the warning belongs to them. If you use one from another manufacturer, ask them to confirm how it complies.
Examples in a small business
- The website chatbot: a first message like «I am an artificial intelligence assistant» and, even better, a visible label throughout the conversation.
- The WhatsApp assistant that answers and makes appointments: The same thing, in the first message. And if the conversation passes to a person, let them know.
- Calls with synthetic voice: The voice says it at the beginning; a simple whistle is not enough.
- Videos with an avatar or cloned voice: Visible warning. We apply it at home: the videos we publish with the cloned voice of our founder carry the warning «AI-generated voice from Albert Lens, with his consent» and the AI content label from TikTok and YouTube.
Timelines and penalties
- Technical marking: The manufacturers of content-generating systems mark it in a machine-readable format. If the system was already on the market before August 2, they have until then to comply with the new requirements. December 2, 2026 (European Commission and AESIA).
- Aid: one Code of good practices volunteer (June 10, 2026) and some others EU icons Free for labeling AI content. The icons are optional; the warning is not.
- Fines: up to 15 million euros, or, if it is a company, 3% % of its total global turnover for the previous year, whichever is greater; for SMEs, the lesser of the two (Article 99, paragraphs 4 and 6). In Spain, the Draft organic law The law that sets the sanctions has been in Congress since June 12, 2026.
How we do it
For us, compliance starts from the design, not at the end. The assistants we deploy are presented as AI; each person enters with their permissions and everything is recorded. Our AI never decides on its own: it proposes, and a person always makes the final decision. That’s what we call Human in the Loop.
Forrellat Consultors, our data protection partner and Data Protection Officer registered with the AEPD, reviews the data protection documentation for each project before signing it. More in Compliance and in AI without government.
If your company already serves customers with AI, or is considering it, we can review it with you.
Frequently Asked Questions
Do I have to notify you if my website's chatbot uses artificial intelligence?
Yes, unless it is obvious. As of August 2, 2026, Article 50 of the European AI Regulation requires that anyone who interacts with an AI must be aware of this from the very first interaction. The simplest way: the chatbot should present itself as an AI assistant in its first message.
What is a deepfake according to the AI Regulation?
An image, audio or video generated or manipulated with AI that resembles real people, places or events and may appear authentic, such as a cloned voice. Whoever publishes it must state that it was created with AI.
What penalties are there for not complying with Article 50?
Up to 15 million euros, or 3% % of the total global turnover of the previous year, whichever is higher; for SMEs, the lower of the two. In Spain, the law that concretizes the sanctions is currently being reviewed by Congress.