As of August 27, 2026. Updated to reflect the European Commission final guidelines on Article 50 of July 20, 2026 and the entry into force of Digital Omnibus Regulation (EU) 2026/1744 on July 27, 2026.
The EU AI Regulation has already faced repeated criticism. But a new threat now looms for creatives and businesses: cease-and-desist letters under competition law due to the lack of or insufficient labeling of AI-generated content. This is because the labeling requirement for AI content, which takes effect on August 2, 2026, is likely to be classified as a rule of market conduct within the meaning of Section 3a of the German Unfair Competition Act (UWG).
The underestimated danger: Section 3a UWG makes labeling obligations subject to warning
Creative professionals and companies should not underestimate the labeling obligation under Art. 50 (4) of the AI Regulation. It not only protects abstract legal interests, but also aims to create transparency, protect against deception and enable informed decisions.
Who can issue warnings – and what is the threat?
This objective clearly suggests that the provision should be classified as a market conduct rule. The practical consequences are considerable. Not only can competing companies claim injunctive relief and damages in the event of infringements, but competition and consumer associations are also entitled to issue warnings.
The Wettbewerbszentrale has also already published a guideline – numerous proceedings are to be expected.
In contrast to fines imposed by the authorities, companies must expect quick reactions here – similar to the influencer labeling at the time.
When did these rules take effect? The difference between operators and providers
The transparency requirements of the AI Regulation have been in effect since August 2, 2026. There is no transition period for the disclosure of deceptively realistic content under Art. 50(4) of the AI Regulation.
This obligation is addressed to the deployer of the AI system. Under the final guidelines of the European Commission of July 20, 2026, that is whoever decides on the use of the AI system and on how its output is used. Publication alone is not the deciding factor. This is more than a matter of wording: in commissioned work it determines who carries the obligation at all.
The situation is different with regard to the provider’s obligation under Article 50(2) of the AI Regulation, which requires synthetic content to be marked as machine-readable at the time of creation. For systems placed on the market before August 2, 2026, Digital Omnibus Regulation (EU) 2026/1744, in force since July 27, 2026, provides for a four-month transition period, meaning that this labeling is not required until December 2, 2026.
However, those who use third-party AI tools cannot rely on this, because the obligation to label content applies regardless of whether the system being used already technically labels the content. It is important to note that the provider’s machine-readable labeling and the labeling visible to humans are two different things, so one does not replace the other.
What exactly must be labeled?
The labeling obligation concerns so-called deepfakes: image, sound or video content generated or edited by AI that resembles real people, objects, places or events and could be mistakenly perceived as authentic. Both elements must be present, and the Commission guidelines make clear that the percentage of AI involved is irrelevant.
Both directions follow from this: a fully generated but clearly stylized motif may require no label, because no one attributes authenticity to it. Conversely, a small change that is decisive for the message of a photo understood as authentic may cross the threshold. Purely technical post-processing such as color correction, cutting out a subject, or rescaling generally stays below it. For work that is evidently artistic, creative, satirical, or fictional, a disclosure that does not interfere with the effect of the work is sufficient.
A separate, narrower standard applies to text: it must be disclosed only where it informs the public on matters of public interest. Pure product and advertising copy therefore generally falls outside the scope. Where a text does touch on such matters, the obligation does not apply if a person has reviewed it editorially and holds responsibility for it. This exception applies to text alone; for image, audio, and video, editorial review does not help.
Practical example: When a home is too good to be true
A real estate agent is advertising a single-family home on an online real estate portal. The listing features impressive photos: a flawlessly plastered facade, modern windows, and a freshly paved driveway. The interior is just as appealing—bright rooms with immaculate flooring and freshly painted walls.
The estate agent has actually reworked the original photos using AI tools. The real property shows clear signs of wear and tear: the façade is weathered, the window frames are outdated, the flooring is worn and the wall paint is stained. The AI-optimized representations are not labeled at all.
A competitor is aware of the actual circumstances and issues a warning – both for violation of Section 3a UWG in conjunction with Art. 50 Para. 4 KI-VO and for misleading according to Section 5 Para. 2 No. 1 UWG. From the warning party’s point of view, the chances of success are good.
How do you label AI content in a legally compliant way?
The Wettbewerbszentrale published a free guide on this on February 4, 2026. It provides practical guidance on labeling.
The most important labeling rules
- Text-based labeling: formulations such as “AI-generated” or “Created with AI” can be considered.
- Image-based labeling: clear symbols such as pictograms can be considered.
- Audio-based labeling: Acoustic cues immediately before and after the AI-generated section may be used.
It is also important to note that the accessibility requirements of Art. 50 (5) of the AI Regulation must be met.
Overall, the legal requirements therefore leave plenty of scope for concrete implementation. Nevertheless, in case of doubt, more information should be provided rather than too little. The decisive factor is whether viewers could mistakenly believe the content to be authentic.
Special attention: Purchased content
Many companies use stock databases for image material. Here too, the responsibility for labeling remains with the company using the material. It must therefore be checked whether acquired materials are AI-generated. Some image databases already offer filter options for AI content.
The situation differs where an agency or service provider is commissioned. A company that merely orders a creative and leaves the use of AI entirely to the contractor is generally not a deployer; the contractor then owes the disclosure. If the client specifies whether and how AI is used, or signs off on the finished creative, both can be deployers side by side. Who handles the labeling, and who bears the risk internally, therefore belongs in the contract.
Act now: Checklist for companies
This requirement is already in effect. Anyone who has not yet reviewed their own content should systematically check the following areas:
- Website: All embedded images, videos and audio files
- Social media: All channels such as Instagram, Facebook, LinkedIn, YouTube
- Advertising media: print and online ads, newsletter
- E-commerce: product images and videos in online stores
- External content: Purchased stock materials and their origin
In addition to deepfake labeling, the Wettbewerbszentrale’s guide also deals with related topics such as misleading “AI washing” or the labeling of chatbots. It offers an initial sound orientation.
What Companies Should Be Reviewing Now
Anyone who underestimates the labeling requirement risks receiving costly cease-and-desist letters from competitors or industry associations starting in August 2026. Case law will only develop specific standards over time—until then, a cautious and strict interpretation of the labeling requirements is recommended.
Creative professionals and companies are well advised to adapt their communication at an early stage. This is because, unlike official proceedings, warnings from competitors can be issued very quickly.
Frequently asked questions
Does AI-generated content have to be labeled since August 2026?
Yes. For certain AI-generated content, particularly hyper-realistic images, videos, and audio files, the transparency requirement under the EU AI Regulation has been in effect since August 2, 2026.
Can a failure to include an AI label result in a warning?
Yes. The labeling requirement is likely a rule of market conduct within the meaning of Section 3a of the Unfair Competition Act (UWG). Violations are then subject to cease-and-desist letters under competition law.
Who may issue a warning, and what is at stake?
Competitors as well as qualified trade and consumer associations. Injunctive relief and cost claims may follow.
Who does the labeling obligation apply to?
The deployer of the AI system, meaning whoever decides on the use of the AI and on how its output is used. Who ultimately publishes the content is not decisive on its own.
Who labels content created by an agency?
If the agency decides on the use of AI itself, the obligation is the agency’s. If the client sets requirements as to whether and how AI is used, or signs off on the creative, both can be deployers side by side. The allocation should be set out in the contract.
Is there a transition period for AI labeling?
Not for deployers: the obligation under Art. 50(4) of the AI Regulation has applied since August 2, 2026. Only providers have until December 2, 2026 to apply machine-readable marking to systems placed on the market before that date.
New labeling requirements—such as those for AI content—carry the risk of receiving cease-and-desist letters. We can help you ensure your public presence complies with the law—learn more about competition law.
