With the AI Act’s transparency obligations now in effect, you’re suddenly seeing notices everywhere: An image carries a small watermark. A chatbot greets you by saying that it isn’t a human. One blog article has a footer, another doesn’t. Some websites have started adding notices to many kinds of content as a precaution. Others haven’t changed anything at all.
This inconsistency feels like a contradiction. Either there’s an obligation, or there isn’t. In reality, the difference isn’t due to one side being more careful and the other being less so. It exists because two different situations can look the same from the outside.
Transparency is not the same as labeling
In everyday language, the two words are often used interchangeably. In the AI Act, they represent two different ideas.
Transparency means that a person should know what they’re interacting with. If a chatbot responds, it should be clear that there isn’t a person on the other end. That’s a statement about the situation, not about the content.
Labeling means that a specific piece of content carries a notice about its origin. An image, a video, or a text receives a label because it was artificially generated or modified. That’s a statement about the product, not about the conversation.
An example makes the distinction easier to grasp. If you’re talking to a customer service chatbot, you receive a transparency notice: No person is responding here. The chatbot’s individual response doesn’t need an additional label because of that. While the two obligations overlap in some situations, they are fundamentally separate.
What the AI Act is actually trying to achieve
The AI Act doesn’t count labels because labels are valuable in themselves. The goal is something else: No one should be deceived by content without having the opportunity to recognize that deception.
One example makes this clear—a case that isn’t controversial. A video shows a well-known person saying something they never actually said. Without a notice, that video can influence real decisions, such as during an election or in the middle of a conflict. Those are exactly the kinds of situations the rules are intended to prevent.
A text that was created with AI assistance but reviewed and editorially approved by a human does not create the same kind of risk of deception. The responsible person stands behind the content, regardless of which tools helped produce it.
That’s why the AI Act distinguishes between different situations
That objective is why not every AI system, every text, every image, and every chatbot is treated the same way. The AI Act distinguishes between ProviderUnder the AI Act, the role of the person or organization that develops an AI system and places it on the market under its own name or trademark
Auch bekannt als: AI Act Provider, who develop or make AI systems available, and DeployerUnder the AI Act, the role of the person or organization that professionally uses an AI system under its own responsibility
Auch bekannt als: AI Act Deployer, who use them. Which obligations apply depends on that role and on the specific use case.
What that means in everyday life
For most everyday situations, the result is far fewer obligations than the headlines suggest.
If you write a blog article with AI assistance but read it yourself, review it, and take responsibility for it, you are not automatically publishing a text that requires labeling. Editorial responsibility remains with a human, not with the tool.
If you use ChatGPT or another language model for your own notes, emails, or internal documents, the transparency obligations do not apply at all. Those obligations begin when content is made available to the public, not during private or internal use.
If you create images with AI and clearly use them as illustrations or works of art, the starting point is different from someone presenting a photorealistic image as if it were a genuine photograph without any notice. That’s exactly the distinction where the labeling requirement for deepfakes applies—that is, for deceptively realistic fake images, audio, or video.
An automated customer service chatbot with no indication whatsoever of its nature is the kind of situation the transparency obligation was originally designed for.
For anyone who wants to explore further
If you’d like to understand the individual terms in more detail, the glossary includes, among others:
- AI ActEU regulation governing AI systems according to risk levels - in force since August 2024, deadlines for High-Risk AI postponed to 2027/2028 by the Digital Omnibus
Auch bekannt als: EU AI Act, AI Regulation, Artificial Intelligence Act - Transparency ObligationAI Act requirement that certain AI systems or their outputs must be recognizable as such to humans
Auch bekannt als: Transparency Obligations, Disclosure Obligation - ProviderUnder the AI Act, the role of the person or organization that develops an AI system and places it on the market under its own name or trademark
Auch bekannt als: AI Act Provider - DeployerUnder the AI Act, the role of the person or organization that professionally uses an AI system under its own responsibility
Auch bekannt als: AI Act Deployer
