Since 2 August 2026, the transparency obligations under Article 50 of the AI Act have applied. For many users of AI, this raises the question: Does this mean that all content created using ChatGPT or another AI model must now be labelled as AI-generated?

 

The short answer is: No.

 

But where is the line drawn, and when must content be labelled or disclosed under Article 50 of the AI Act?

 

As a simplified rule of thumb: The more strongly AI-generated or manipulated content gives the viewer the impression that it authentically depicts a real person, a real object, a real place or an actual event, the more relevant the transparency obligations become.

 

For example, if a company operates an AI chatbot on its website, users must be clearly informed, at the latest at the beginning of the interaction, that they are communicating with an AI system. A notice such as “You are communicating with an AI-powered chatbot. Responses are generated automatically.” may be an appropriate way of providing this transparency.

 

Another example is a realistic AI-generated video of an employee used for product explanations or training videos, even though the employee never actually appeared in front of the camera for that recording. If the depiction is so authentic that a viewer could assume it is a genuine recording, it may constitute a deepfake within the meaning of the AI Act. In this case, it must be disclosed that the content has been artificially generated or manipulated. In principle, this also applies where the employee has consented to the use of their digital likeness.

 

The situation is different, for example, in the case of an operating manual created with the assistance of AI. The use of AI alone does not, as a general rule, trigger a visible labelling requirement under Article 50. The specific disclosure obligation for text applies in particular to publications intended to inform the public about matters of public interest. If such a text has also been reviewed or editorially controlled by a human and a natural or legal person assumes editorial responsibility, Article 50 provides for an exception.

 

However, one point remains important: Even where no labelling obligation applies, responsibility for the content remains with the company. AI-generated content should therefore be professionally reviewed, particularly where it concerns technical, legal or safety-related information.

 

Not every form of AI-assisted image editing is automatically subject to a labelling requirement either. Adjustments to brightness, sharpness or contrast, as well as other ordinary technical edits, do not in themselves generally give rise to a labelling obligation.

 

The situation may be different where AI is used to create a new, realistically depicted situation that never actually occurred. For example, if a real company building is depicted in a highly realistic manner as having been destroyed by an explosion, the image may qualify as a deepfake if viewers could mistake it for an authentic image of a real event.

 

Labelling may also be required for artistic, creative, satirical or fictional AI-generated content. However, less stringent requirements apply in such cases: the disclosure may be made in a more discreet manner where it is already clear to the viewer that the content is not an authentic depiction.

 

AI systems are becoming increasingly powerful. They are capable of generating ever more realistic images, videos and voices and can convincingly imitate human writing and communication styles.

 

This is precisely why Article 50 of the AI Act pursues a central objective: People should be able to recognise when they are interacting with an AI system or are confronted with artificially generated or manipulated content.