For most people, the EU AI Act sounds like a distant Brussels document full of risk categories and compliance deadlines. The transparency chapter is different. You can see it working — in social media feeds, in the chatbot window on a company website, in the byline of an AI-assisted article. The new obligations translate into plain, everyday language: machines must say they are machines, artificial content must be labelled, and people must not be tricked into believing a fake is real.
What you should now look for
The AI Act's transparency chapter is built around three practical duties. None of them requires technical knowledge to understand.
Chatbots must introduce themselves. If an AI system interacts with you in text or voice, the provider and the company deploying it have to make sure you know you are talking to a machine. That ChatGPT already says "I am an AI" is now more than politeness — it is a legal requirement across all 27 member states.
Deepfakes need disclosure. Anyone who uses a system to generate or manipulate realistic image, audio or video content of a person, object or event must clearly disclose that the content is artificial. The rule covers both businesses and individuals, though the law makes an exception for evidently creative, satirical, artistic or fictional works — there the disclosure requirement is lighter, so it does not ruin a Monty Python-style parody.
AI-generated public-interest text must carry a machine-readable label. This is the rule that matters most for media and politics. News, official statements and public warnings produced by AI have to carry a label that machines can recognise. There is one important escape: if a human has reviewed the content and takes editorial responsibility, the label is not required. That was a deliberate carve-out for professional journalism.
| Obligation | What it means in practice |
|---|---|
| AI chatbot or voice assistant | Users must be told they are interacting with a machine; most commercial chatbot platforms already do this by default. |
| Realistic synthetic image, audio or video | Deepfakes must be disclosed; lighter rules apply to evidently creative, satirical or artistic work. |
| AI-generated public-interest text | Must be machine-readable labelled unless a human takes editorial responsibility. |
Why this is more than a platform policy
The transparency obligations are the second wave of the AI Act's application. The regulation banned the most dangerous practices — social scoring, manipulative subliminal techniques, exploitative use of people's vulnerabilities — back in February 2025. Now come the rules that govern the everyday content we all encounter.
In a way, the law is catching up with what platforms already started doing voluntarily. OpenAI embeds C2PA metadata in images generated by DALL·E and ChatGPT, Google applies SynthID watermarks to Gemini images, and Meta has been attaching "AI info" labels to posts on Facebook and Instagram. What the AI Act adds is a uniform legal baseline: a company that skips transparent labelling is not just breaking a platform policy, it is breaking European law.
Who enforces it, and what the fines mean
Enforcement is in the hands of national authorities. In the Czech Republic, for example, the Digital and Information Agency coordinates AI Act supervision. Each member state has a designated body, and the European Commission's AI Office watches over the bigger picture.
The fines are substantial. Violating the transparency obligations can cost a company up to €15 million or 3 percent of global annual turnover, whichever is higher. Compare that with the penalties for banned practices — up to €35 million or 7 percent — and you see that Brussels treats misleading people seriously, though it reserves its heaviest artillery for genuinely harmful uses.
What changes in your daily feed, your work and your small business
For the rest of us, the practical consequences are both modest and meaningful. If you run a small e-shop with a customer-service chatbot, the software should now explicitly tell customers they are chatting with an AI. That is a one-line change for most chatbot platforms, but it protects you from an uncomfortable conversation with a regulator.
If you are a content creator, labelling AI-made video or imagery is no longer a matter of platform goodwill. And the next time a spectacular video of a politician circulates, the presence or absence of a label gives you a first — though not foolproof — clue about its authenticity. The European Commission describes this as giving citizens the "right to know" when content is synthetic.
There is a second layer under all of it: GDPR. A deepfake of a real person can also violate data-protection law, defamation rules and personality rights. The AI Act labels the content; the other laws protect the person. Both work alongside each other.
To be honest, the rules will not stop determined bad actors. Labels can be stripped, watermarks can be cropped, and the people producing political disinformation rarely care about EU fines. What the transparency obligations do is shift the burden of proof: the default becomes "show me the label", not "trust me, it's real". That is a quiet but real change in European digital culture.
For readers who want to understand how to spot manipulated content in practice, our magazine coverage has been tracking the AI Act throughout its rollout. The regulation's story is far from over — the stricter high-risk rules follow in 2026 and beyond.
If I post a funny AI video of my dog on Instagram, do I have to label it?
The AI Act targets users who generate or manipulate realistic content, but it allows lighter treatment for evidently creative, satirical or artistic work — and enforcement authorities are unlikely to chase harmless personal jokes. The risk grows dramatically when realistic fakes show real people, because then defamation and GDPR rules can apply on top of the AI Act.
Does my company's website chatbot really need to say it is a bot?
Yes. If the chatbot interacts with customers or the public, users must be informed they are interacting with an AI system. Most commercial chatbot platforms already include this disclosure by default, but it is worth checking your setup — the fine for ignoring it can reach millions of euros.
Will I notice the labels on AI-generated images in my social media feed?
Sometimes. Some labels are visible — like a small "AI info" badge under a post — while others are embedded invisibly in image metadata. The EU rules require disclosure, but they do not dictate exactly how visible every label must be, which is why the experience varies between platforms.
