Since 2 August, the transparency rules under the European AI Act have officially been in force. No more vague talk about the future, but a law you already need to comply with today. If you fail to do so, you could face a fine of up to €15 million or three percent of your global turnover. Time to take a closer look at what this actually means for your organisation.
A short timeline
The AI Act as it has now entered into force is not actually new. The law already entered into force on 1 August 2024, but has been rolled out in phases ever since. First came the bans on certain AI practices, followed by the rules on AI literacy within organisations. Since 2 August 2026, the next and much broader wave has taken effect. Not only the transparency rules under Article 50, but also effective enforcement at national and European level now starts. Companies can therefore actually be inspected and fined from now on. The full rollout of the law continues until 2 August 2028, so this is not the end point. It is, however, the moment when the law starts to have real teeth.
What has actually changed
Article 50 of the AI Act revolves around one principle. People need to know when they are dealing with AI. In practice, this means three things:
Use a chatbot? Then you must make it clear from the first interaction that the user is talking to AI, not a person.
Create realistic AI content, images, audio or video that could be mistaken for real content? It must be visibly or audibly labelled as artificial. Think of an AI generated product image for a campaign, a synthetic voice in a voice over, or an AI avatar in an advertising video. Video and audio in particular are the hardest content types to label and track consistently, so that is where you face the greatest risk.
Publish AI generated text on a matter of public interest? Then a label is required, unless a human performs the final editorial review and takes final responsibility. That is also your easiest way around the labelling requirement.
The rules apply to companies, public authorities, self employed professionals and professional influencers. Purely private use is excluded.
How much does such a label actually need to say?
Legally, the key point is this: it must be clear and unambiguous to an average, attentive user, and this must be the case from the first interaction. A tiny label in the corner of an image or a vague mention in the terms and conditions is not enough. But a detailed description of exactly what was edited or generated is not legally required. A simple label such as ‘Edited with AI’ can therefore be sufficient, as long as it is clearly visible and unambiguous.
One nuance: in June 2026, the European Commission published its Code of Practice on Transparency of AI generated Content. It is voluntary, but serves as a practical benchmark and distinguishes between content that was fully generated by AI and content that was AI assisted. Each has its own recommended label. This is something to take into account now when setting up a labelling system.
Why this can’t be left on the shelf
A fair question, because if nobody is watching, a law can feel optional. In practice, enforcement takes place through national supervisory authorities in each member state, supported by European coordination for cross border cases. The problem in Belgium is that this national supervisory authority has not yet been appointed. That does not mean there will be no enforcement. It simply means the rules may become clear faster than the authorities responsible for enforcing them. For your organisation, this changes little. The obligation applies today, regardless of whether the supervisory authority is already operating at full capacity.
5 concrete steps you can take
Map where AI is being used.
Chatbots, content generation, image editing, automated copy. Create a list for each team. Not just what marketing uses, but also what sales, HR or customer service teams may already be using on the side.
Check your output.
Which content that customers or visitors see has been created with the help of AI or generated by AI? That content needs a label. Think blog articles, social posts, generated product images and videos with synthetic voices.
Put final editorial review in place.
Human oversight of AI generated text on matters of public interest can save you from having to add a label. Make sure that review actually happens, not just on paper. Who reads it, who approves it, and where is that documented?
Build governance into your systems, not on top of them.
A standalone policy document that nobody reads solves nothing. AI governance works when it becomes part of how your systems are built, rather than an extra step that everyone skips when a deadline is approaching.
Make sure you know what is in your content.
And this is often where things get difficult. Before you can label something, you need to know it exists. In practice, many organisations do not have a complete overview of which content is AI generated, where it is stored and who uses it. Assets disappear into folders, DAM systems or old campaigns, without anyone knowing what they contain or how they were created.
Who actually checks this?
A fair question, because if nobody is watching, a law can feel optional. In practice, enforcement takes place through national supervisory authorities in each member state, supported by European coordination for cross border cases. The problem in Belgium is that this national supervisory authority has not yet been appointed. That does not mean there will be no enforcement. It simply means the rules may become clear faster than the authorities responsible for enforcing them. For your organisation, this changes little. The obligation applies today, regardless of whether the supervisory authority is already operating at full capacity.
The blind spot behind the AI Act
This is the paradox. Companies are being asked to be transparent about AI content, while those same companies often do not even have a clear view of their own content. Not because they are careless, but because traditional systems simply were not built for this. Tags do not capture what is actually inside a file. Videos and audio often get one label for the entire file. Nobody knows three clicks further which part was AI generated and which part was not. You can hardly comply with a labelling requirement for content you cannot find yourself.It’s hard to comply with a labeling obligation for content you can’t even find.
Compliance starts with visibility
This is where your biggest exercise lies. AI Act compliance is not just a legal exercise. It is a good reason to review your entire content landscape. What content do you have, where is it stored, how was it created and who still has access to it? Companies that do this exercise now solve two problems at once. They comply with the AI Act, and they finally know what they have.
Not immediately excited about this? We completely understand. It can be a substantial task, and you are rarely sure whether you have really mapped everything. A comprehensive content scan by our experienced team can help you get started faster and more thoroughly. In a short time, you get a precise view of how your content is currently created, tagged, stored and retrieved. You immediately see where the AI generated content is that needs a label, and where you are already dealing with hidden costs and duplicated work.
Not slowing down, just being deliberate
The AI Act is not there to slow down AI adoption. It is about trust. Just as the GDPR provided a framework for personal data, the AI Act provides a framework for trustworthy AI. Companies that take this seriously now can build an advantage over those who wait until the first fine arrives.
Curious about where your organisation stands today, both in terms of AI transparency and content visibility? Request a content scan and we will map it out together.
Win the
battle of the brands
Start with automation of visual content
Ybe Jacobs
Win the
battle of the brands
Start met visuele content automation
