Labelling Obligations for AI Content: Transparency Obligations under the EU AI Act
(As of: June 2026)
AI has long since ceased to be an internal-only tool and is now visibly used externally: chatbots in customer contact, AI-generated images in campaigns, synthetic voices, automatically generated texts. This raises a question that long remained open in practice and is now being answered concretely: Do AI-generated contents have to be labelled as such at all?
The short answer is: Yes. From 2 August 2026, this obligation in principle affects almost every company that uses AI systems in the course of its professional activity.
The legal framework: Which compliance rules exist in general?
Labelling obligations for AI content do not arise from a single provision but, in particular, from various legal sources:
- Art. 50 of the AI Act (AI Regulation; EU AI Act) establishes general transparency obligations for the use of certain AI systems and – unlike the other regimes – addresses almost every company.
- Art. 35(1)(k) DSA obliges providers of very large online platforms and search engines to label manipulated image, audio and video content.
Unfair competition law (UWG) is in practice the most important flanking protection – though not automatically via the breach-of-law route: Whether Art. 50 AI Act is a market conduct rule within the meaning of Section 3a UWG is as yet unresolved. A robust line of attack can run via misleading practices – as the withholding of material information (Section 5a UWG) or, for instance in the case of supposedly genuine "customer testimonials", as active misleading (Section 5 UWG) or via the blacklist (Annex to Section 3(3) UWG). In these cases, competitors can issue cease-and-desist letters.
Further requirements may apply in addition, in particular data protection information obligations (Art. 13/14, Art. 22 GDPR) as well as personality rights in the case of deepfakes of real persons; sector-specific rules, for instance in pharmaceutical advertising or capital markets law, come on top.
Art. 50 AI Act at a glance
The AI Act entered into force on 1 August 2024. Its practically central provisions, including the labelling obligations of Art. 50 AI Act, take effect from 2 August 2026 (Art. 113 sentence 2 AI Act). The Regulation's declared aim is to make AI systems transparent, reliable and safe and thus to ensure a high level of protection.
Its scope extends to the entire EU, and no national implementation is required. The obligations apply directly.
In substance, Art. 50 AI Act establishes graduated transparency obligations that are distributed between providers and deployers. Providers must disclose that persons are interacting with an AI system (para. 1) and mark AI-generated content as artificially generated in a machine-readable format (para. 2). Deployers must disclose the use of emotion recognition and biometric categorisation systems (para. 3) and label deepfakes and AI-generated texts on matters of public interest as artificially generated or manipulated (para. 4). What this labelling must look like in concrete terms is explained in the following articles.
A temporal differentiation is introduced by the Digital Omnibus, politically agreed in May 2026: For the machine-readable marking under Art. 50(2) AI Act, providers of generative AI systems already placed on the market before 2 August 2026 receive a transitional period until 2 December 2026. The remaining transparency obligations of Art. 50 AI Act – and thus in particular the deployer obligations in focus here – remain set at 2 August 2026. However, these relaxations only become legally effective upon formal adoption and publication in the Official Journal. Until then, the cut-off date applies without restriction.
Who must label? Providers and deployers
Who bears which obligation is decided by a distinction that shapes the entire system of obligations under the AI Act: the distinction between provider and deployer.
A provider under Art. 3 No. 3 AI Act is whoever develops an AI system or has it developed and places it on the market under its own name or trademark. The provider bears the primary regulatory responsibility.
A deployer under Art. 3 No. 4 AI Act is whoever uses an AI system in the course of a professional activity.
Crucially: The assessment is functional, not based on the contractual designation of the parties. Whoever poses contractually as a mere user but actually acts like a provider will be held accountable accordingly.
Which content must be labelled?
The labelling obligations cover a broad spectrum of AI-generated content. Depending on the type of content, the obligation falls on the provider, the deployer or both:
- AI avatars and chatbots – where persons interact with an AI system, this must be recognisable (disclosure by the provider, Art. 50(1) AI Act).
- Texts of public relevance – machine-readable marking of AI-generated texts by the provider (para. 2); additionally, labelling of AI-generated texts on matters of public interest by the deployer (para. 4).
- Images, audio and video, so-called deepfakes – machine-readable marking as artificially generated by the provider (para. 2) and – insofar as deepfakes are concerned – labelling by the deployer (para. 4).
Each of these categories has its own thresholds and requirements – such as the question of when an edit is subject to labelling at all and how the notice must be placed in concrete terms.
A detailed overview of how deepfakes must be labelled can be found in our article "How must deepfakes be labelled?"
What are the consequences of infringements – and what if a competitor fails to label?
The labelling obligations of Art. 50 AI Act are not only a regulatory compliance programme but also a risk factor under competition law.
In regulatory terms, market surveillance measures and fines loom.
In addition, private enforcement via the UWG comes into consideration. Whether Art. 50 AI Act itself constitutes a market conduct rule within the meaning of Section 3a UWG remains unresolved.
Conclusion and outlook
The labelling obligations of the AI Act are not a distant regulatory threat but, from 2 August 2026, applicable, directly effective law – with no room for national implementation. A grace period exists only within narrow limits, namely for the machine-readable marking of systems already on the market before the cut-off date. Otherwise, practically every company that uses AI-generated content externally is affected.
The first step is therefore an honest stocktake: Where in the company are AI-generated or AI-manipulated contents created, and in which role – as deployer or, exceptionally, as provider – does one act in doing so? Those who clarify these questions early can build the concrete labelling requirements into their processes in an orderly manner instead of catching up under time pressure.
*You want to know which of your AI applications will be subject to labelling from August 2026 and how to set up your processes in a legally compliant way? We analyse your concrete use of AI systems, classify your role as deployer or provider and develop a practicable compliance line. And if you have received a cease-and-desist letter or observe that a competitor is not labelling AI content: We also assist you in enforcing and defending against claims under competition law.
