Since 2 August 2026, Article 50 of the EU AI Act applies. If your business uses AI, you mainly have to disclose deepfakes, meaning AI images, videos or audio that look real, and AI texts on matters of public interest that no person has reviewed editorially. The invisible, machine-readable marking of AI content, on the other hand, is the job of providers such as OpenAI or Google, with a deadline of 2 December 2026 for systems that were already on the market before.
I am a developer, not a lawyer, and this is not legal advice, just the operator's perspective, under EU and German law. My fourteen AI employees run on Claude Team, some also on open models or through the API. I read the text of the regulation on 27 October 2026 in its current consolidated version on EUR-Lex, which includes the changes made by the Digital Omnibus Regulation on AI, Regulation (EU) 2026/1744. More on AI in day-to-day business is collected on my AI Basics page.
What Article 50 covers
The regulation distinguishes two roles. Providers develop an AI system and place it on the market under their own name, such as OpenAI with ChatGPT. Deployers use an AI system under their own authority, in a professional context. If your business uses ChatGPT, Claude or an image generator, you are a deployer.
The key sentences, word for word
For users in small businesses, paragraphs 4 and 5 matter most. This is the current wording:
"Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated." (Art. 50(4), first subparagraph, first sentence)
"Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated." (Art. 50(4), second subparagraph, first sentence)
This text obligation does not apply "where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content".
"The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure." (Art. 50(5), first sentence)
Article 3(60) defines a deep fake as AI-generated or manipulated image, audio or video content "that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful".
What applies from when
- Since 2 August 2026: Article 50 applies to providers and deployers. The Digital Omnibus Regulation did not postpone this date. Within Article 50 it only rewrote paragraph 7, which governs how the Commission assesses codes of practice, meaning voluntary rulebooks for implementation.
- Until 2 December 2026: a transition period only for the machine-readable marking under paragraph 2, and only for AI systems placed on the market before 2 August 2026. The Omnibus Regulation added this deadline as Article 111(4). It changes nothing about your own duty for deepfakes.
- Older content: According to the European Commission, content generated before 2 August 2026 does not need to be labelled retroactively. The Commission does encourage it, though.
- From 2 December 2026: AI systems that, under certain conditions, generate realistic intimate or sexual depictions of an identifiable person without their explicit consent become prohibited (Art. 5(1), point (ba), added by the Omnibus Regulation). No label helps here.
What this means day to day
The following cases are typical examples, not an assessment of your situation:
The European Commission makes clear in its answers on Article 50 that purely formal checks are not enough: correcting spelling and grammar does not count as human review. For a chatbot, paragraph 5 requires the notice at the latest at the first interaction. In practice, that means the first sentence. What that looks like on the phone is in AI Receptionist: Providers, Pricing, Law.
And private use? Anyone using AI purely privately is not a deployer, and the regulation does not apply. The Commission adds, however, that anyone who gains an economic benefit from it on a regular basis is acting professionally and counts as a deployer. That covers, for example, freelancers who win customers through social media.
How to label correctly
According to the European Commission, the disclosure must be perceivable by people without technical tools, for example as a visible or audible label. You expressly cannot rely on the provider's invisible mark. A short notice right next to the content is often enough, such as "Image generated with AI" or "Voice generated with AI". For evidently artistic, satirical or fictional works, paragraph 4 only requires disclosure in an appropriate manner that does not hamper the enjoyment of the work. Under paragraph 5, the notice must also be accessible. A notice that only appears inside an image therefore also belongs in the alt text or the caption.
Platforms such as Instagram and Facebook also apply their own labels. What happens there is in Can AI Create Instagram Reels?, what is embedded in generated images is in Creating AI Images: Free and Free to Use, and what applies to AI video is in AI Video Generator vs. AI Video Editing.
What happens in case of a breach
Article 99(4) provides for fines of up to 15 million euros or 3 percent of worldwide annual turnover for breaches of the transparency obligations in Article 50. For small and medium-sized enterprises, paragraph 6 applies the lower of the two amounts. These are ceilings: under paragraph 1, member states must take into account the interests of small and medium-sized enterprises and their economic viability when imposing penalties. In Germany, the market surveillance authority, which supervises AI systems on the market, is the Bundesnetzagentur, the Federal Network Agency, under section 2(1) of the German AI implementation act (KI-MIG).
What to watch out for
- When in doubt, label it. A notice costs nothing, deception can cost trust and money.
- Other laws still apply. Paragraph 6 makes clear that other transparency obligations under EU or national law remain unaffected.
- People keep their rights. Regardless of Article 50, deepfakes can violate a person's general right of personality, according to the German Federal Ministry of Justice. With real people, get their consent first.
- A rule for the team. Decide who judges whether content looks real and what the notice looks like.
Frequently asked questions
Do I have to label AI images?
Under Article 50, if they are deepfakes, meaning they resemble real people, places or things and would falsely appear authentic. Clearly artificial illustrations usually don't fall under it. Platform rules can apply on top.
Do I have to label AI-generated text?
Only texts meant to inform the public on matters of public interest, and not even then if a person has reviewed them for content and someone holds editorial responsibility. A reviewed company blog usually does not fall under it.
When does the AI labelling obligation apply?
Since 2 August 2026. Only the providers' machine-readable marking has a deadline of 2 December 2026 for systems that were already on the market before.
Does the disclosure obligation apply to private individuals?
Not for purely private use, because then you are not a deployer within the meaning of the regulation. Anyone who gains an economic benefit from it on a regular basis counts as acting professionally, according to the European Commission.
Do I have to label older AI content retroactively?
No, according to the European Commission, not for content generated before 2 August 2026. The Commission does recommend it where possible.
Where to go from here
Go through your last ten posts and mark everything that was generated with AI and looks real. Then settle on a fixed notice text. How to write this down as a rule for the whole team is in AI Usage Policy for Small Teams. Who owns the AI image afterwards is a different question, and I answer it in the next step: AI and Copyright: Who Owns AI Images?. You will find every AI term explained briefly in my AI glossary. How others label their AI content is what we discuss in my community Claude Practitioners. Membership does not replace a legal review.
