
or two years the EU AI Act was something that would happen later. Since 2 August 2026 a part of it applies to almost every organisation that communicates with the help of AI. It is the part about transparency, and it is much smaller and more reasonable than the headlines suggest.
I am a communication consultant and not a lawyer, so read this as an overview and not as legal advice. But the questions these rules raise are communication questions, and most small teams can answer them in an afternoon.
In summer 2026 the EU changed the timetable. The strict rules for so-called high-risk systems, such as AI that screens job applicants, were moved to December 2027 and August 2028. Many people read the news as “the AI Act has been delayed”.
That is not the case. The transparency rules kept their date. They are what matters for communication teams, because they deal with what people are told when they meet AI.
People should be able to tell when they are dealing with a machine, and when something that looks real was made by one. As a communicator I find it hard to object to that. It is what I would recommend anyway, law or no law.
1. Does anyone talk to a machine on your channels? If you have a chatbot on your website or an automated assistant answering messages, people must be told that they are talking to an AI, at the start and not hidden in the small print. A plain first line is enough: “I'm an automated assistant. I can answer questions about our opening times and events.”
2. Do you publish images, audio or video that look real but are AI-made? A generated photo of an event that never happened, or a voice that sounds like your director, has to be marked as artificially generated. Clearly artistic or satirical work gets a lighter touch, but the label still belongs there. My advice goes a step further: don't use realistic AI images of people in your communication at all. Any trust you lose that way is hard to win back.
3. Do you publish AI-written text on matters of public interest without a person editing it? This is the one that worries people most, and it affects fewer teams than they think. If a person reviews the text and someone takes editorial responsibility for it, you don't have to label it. That is how a careful team works anyway. The rule is aimed at automated publishing with nobody in between.
You don't have to label every email, post or report that began as an AI draft and was then rewritten by a person. And using a chatbot to summarise a document doesn't have to be registered anywhere, nor does an ordinary writing assistant need a risk assessment.
There is one more duty that has applied since February 2025 and is often overlooked: people who use AI at work should know enough to use it sensibly. For a small team, a short training session and a one-page policy cover it.
The fines in the law are written for large companies, and I don't expect an authority to start with the neighbourhood foundation. The better reason to do this is your readers. Being open about where a machine was involved costs you one sentence. If they find out later that you kept quiet, you lose their trust.
If you would like to go through the three checks for your channels, book a free intro call.
