From 2 August 2026, under the European AI Regulation (Regulation (EU) 2024/1689), labelling requirements will apply to image, audio and video deepfakes, as well as to AI-generated text in the public interest. This raises the question for e-commerce: which AI-generated images must be labelled – and which do not need to be?
Article 50(4) of the AI Regulation (Regulation (EU) 2024/1689) obliges operators of AI systems that generate or manipulate deepfakes to disclose that the content has been artificially generated or manipulated.
A deepfake, as defined by the AI Regulation, is AI-generated or AI-manipulated image, audio or video content that resembles existing persons, objects, places or events and would falsely appear to a viewer to be authentic or truthful.
The AI Regulation is structured around different risk spheres, in each of which the specific risks associated with the use of AI are decisive (see Recitals 133 and 134 of Regulation (EU) 2024/1689). Accordingly, the specific risk of a deepfake is said to lie in the deception regarding the authenticity of the content depicted: the viewer is intended to mistake an artificially generated or manipulated representation for a genuine or truthful one. The labelling requirement therefore applies only where such a potential for deception exists in the specific case. If there is no potential for deception, there is also no basis for classification as a deepfake – which is grounded in the legislative purpose – nor for the associated disclosure requirement. Where, however, this risk does exist, the labelling creates transparency and, as a response to the risk, eliminates the danger that the viewer will be misled as to the authenticity of the representation.
The European Commission’s guidelines make it clear: an AI-generated advertising image is a deepfake if it is capable of misleading the viewer as to the actual appearance, characteristics or use of the product. According to the guidelines, however, a genuine product set against a non-misleading AI-generated background is not a deepfake. The specific context of use is decisive.
It remains to be seen whether even truthful AI representations of a product are considered deepfakes and must be labelled as such. There are strong arguments in favour of truthful AI product images not being classified as deepfakes. According to the purpose of the standard, the labelling requirement is intended to protect the viewer from a false attribution of authenticity and the resulting deception. Illustrative backgrounds without a specific location reference should already be classified as generally unproblematic under the existing guidelines. Even in the case of a truthful product representation, this potential for deception is precisely what is lacking. The product is shown as it actually is, so that, in this respect, an AI-generated image is functionally no different from a professional advertising photograph. The content conveyed remains the same – only the medium of production has changed. It therefore stands to reason that one should not assume the presence of a deepfake within the meaning of the AI Regulation, precisely because the viewer is not being deceived about the reality depicted.
According to this view, this means for online retail that truthful product images are generally not deepfakes. This also applies, in principle, to AI-generated models used to showcase clothing, because consumers focus on the product features rather than the authenticity of the model. The situation is different if AI gives the impression that an identifiable public figure is actually promoting a product. In that case, it constitutes a deepfake that must be labelled as such.
The EU Code of Practice stipulates the use of a standardised, high-contrast ‘AI’ icon for AI-generated content. In addition, a short text such as ‘AI-generated’ may be used. Deepfakes must be clearly and unambiguously labelled, without the need for technical aids, at the latest upon the first interaction with the content. For product images, it is therefore recommended that the label be placed directly on the image.
There is no labelling requirement for content that forms part of obviously artistic or satirical works. However, its artificial origin must be disclosed in an appropriate manner without impairing the presentation or the enjoyment of the art.
To implement the new rules, e-commerce companies should, in particular, take the following measures:
(Article 50(4) of Regulation (EU) 2024/1689 – the AI Regulation; Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689 (the ‘AI Act’), 20 July 2026)
Subscribe to our GvW Newsletter here - and we will keep you informed about the latest legal developments!
