As of August 2, 2026, the so-called AI labeling requirement applies throughout the EU: Chatbots, AI-generated images, videos, audio files, and certain texts must be labeled in such a way that people can recognize that artificial intelligence was involved. The legal basis for this is Article 50 of the EU AI Regulation (AI Act). Anyone who violates this regulation risks fines of up to 15 million euros or 3 percent of their global annual revenue.
It sounds clear-cut—but in practice, it isn’t always that simple. We’ll explain who is affected, what exactly must be labeled, where the exceptions lie, and how you can implement labeling in a legally compliant manner.
Key Points at a Glance (TL;DR)
- The AI labeling requirement under Article 50 of the AI Act has been in effect since August 2, 2026.
- The following must be labeled: AI chatbots, AI-generated or AI-edited images/videos/audio (deepfakes), and AI-generated text on topics of public interest.
- This primarily affects providers and operators of AI systems—purely private, non-professional use is largely exempt.
- There are exceptions, such as for tools used solely for editing and for editorially reviewed content.
- A grace period until December 2026 applies to the technical labeling of AI-generated content.
- Violations are punishable by fines of up to 15 million euros or 3% of global annual revenue.
What is the AI labeling requirement under Article 50 of the AI Act?
The AI labeling requirement is part of the EU AI Regulation (Regulation (EU) 2024/1689), which is coming into effect in phases. Article 50 governs transparency requirements—that is, the question of when people must be informed that they are interacting with AI or viewing AI-generated content.
The goal is simple: to curb deception and disinformation and strengthen trust in digital content. “With the start of implementation, we are taking an important step toward AI that people and businesses can understand and trust,” said EU Vice President Henna Virkkunen at the launch.
Important: There is no blanket labeling requirement for every use of AI. Article 50 targets specific risk scenarios. Anyone using ChatGPT to draft an email does not have to disclose this. However, anyone publishing a deceptively realistic deepfake certainly must.
When does the labeling requirement take effect?
The transparency rules have been in effect since August 2, 2026. However, there are transition periods for some of the requirements:
- Technical labeling: For generative AI systems that were already on the market before August 2, 2026, there is a grace period for machine-readable labeling until December 2026 (Article 50, paragraph 2, as amended by the so-called AI Omnibus Package).
- Legacy deepfakes: For deepfakes created before August 2, 2026, there is no retroactive labeling requirement—though it is recommended.
What Must Be Labeled? The Four Scenarios
The EU distinguishes between four scenarios. Two of these concern providers of AI systems (who supply the technology), and two concern operators (who use the AI and publish content).
| Scenario | Who is responsible? | What needs to be done? |
|---|---|---|
| Interaction with an AI system (e.g., chatbot) | Provider | Design the system so that users recognize they are speaking with an AI |
| AI-generated or AI-edited content (image, video, audio, text) | Provider | Mark content as “synthetic” in a machine-readable format and make it recognizable |
| Emotion recognition & biometric categorization | Operator | Inform affected individuals about the use of AI |
| Deepfakes & AI-generated texts on topics of public interest | Operator | Clearly label content as AI-generated |
What specific steps do I need to take as a user, company, or creator?
For most cases relevant in practice, the obligation can be broken down into three areas.
1. Chatbots and AI assistants
If you use an AI chatbot on your website or in your app, users must be able to tell right from the start that they’re not chatting with a human. A note in the fine print isn’t enough—the information must be provided openly and immediately, for example, with a clear “I am an AI assistant” at the start of the conversation. Exceptions apply in cases where it is obvious to any reasonably informed person that the entity is an AI.
2. AI Images, Videos, and Deepfakes
A deepfake is any AI-generated or AI-edited image, audio, or video content that resembles real people, places, objects, or events to such an extent that it could be mistaken for the real thing. Such content must be clearly and conspicuously labeled as AI-generated—regardless of whether there is any intent to deceive.
This is precisely where it’s worth taking a look at just how powerful modern AI image tools have become: from AI-powered photo sorting to systems that turn simple sketches into finished graphics. Google AutoDraw, for example , which generates clean graphics from rough drawings, demonstrates just how quickly this can happen. And we’ve examined how AI automatically tags entire image collections using the example of AI-powered photo organization on the Synology NAS.
3. AI-Generated Texts on Topics of Public Interest
If AI-generated texts are published to inform the public about topics of public interest—such as news—and these texts have not been editorially reviewed, they must be labeled as such. However, if a human is responsible for reviewing the content and assumes responsibility for it, an exception applies. This requirement generally does not apply to AI-generated advertising copy or automated product descriptions.
Exceptions: When You Don’t Need to Label
Not every use of AI triggers a labeling requirement. The most important exceptions:
- Editing tools: If the AI is used only as a support tool—such as for standard image editing, spell-checking, or filters—and does not significantly alter the input data, the labeling requirement does not apply.
- Editorial oversight: AI-generated text that is reviewed and approved by a human does not need to be labeled as AI-generated.
- Private use: The operator obligations apply to professional or commercial use. Purely private, non-professional use is largely exempt.
- Art and Satire: For works that are clearly artistic or satirical, the labeling must be done in a way that does not impair the presentation.
What penalties apply for violations?
National market surveillance authorities are responsible for enforcement; for certain systems, the EU’s AI Office and the European Data Protection Supervisor are also responsible. The fines are substantial:
- up to 15 million euros or up to 3 percent of global annual revenue —whichever amount is higher;
- up to 750,000 euros for EU institutions;
- a proportionality rule for small and medium-sized enterprises (SMEs).
How do I properly label AI content?
There is no legally required standard labeling—what matters is that the notice is clear, unambiguous, and visible. However, the EU provides guidance:
- Commission Guidelines on Article 50: These clarify who falls within the scope of the regulation and how the obligations should be interpreted.
- Code of Practice on Transparency for AI-Generated Content: a voluntary tool (published in June 2026) that allows providers and operators to demonstrate their compliance.
- Official EU Icons: The EU has published three optional symbols that can be used to label AI-generated content.
In practice, in many cases, a visible label such as “AI-generated” on the image, in the video, or below the text—as well as, for providers, a machine-readable tag in the file’s metadata—is sufficient.
Conclusion: Transparency Becomes Mandatory—With Room for Interpretation
Starting in August 2026, the AI labeling requirement will bring noticeably greater transparency to chatbots, AI-generated images, and automated texts. For companies and creators, this means, above all, checking where AI is involved and clearly labeling those instances where people might otherwise be misled. The gray areas lie in the details—such as the question of what exactly “substantially altered” or “of public interest” means. If in doubt, it’s best to label the content to be on the safe side. The European Commission’s guidelines and the Code of Practice provide the necessary guidance.
Sources: European Commission – Quick Facts: Transparency rules for AI systems, EU Guidelines on Transparency for AI-Generated Content
