The Kids Top 20 tv show is back. With a new presenter. One problem: she doesn’t exist. She has been entirely generated by artificial intelligence, looks perfect, is smartly dressed and has a physique that – even if she were a real person – does look artificial in places.
The accompanying NOS article raises the question of whether this is simply allowed. As a provider, are you required to disclose that your presenter is AI-generated? The answer is provided below, discussing the requirements set by the AI Act in this context.
The AI Act: a duty to disclose for an AI character?
Let’s start with the most obvious question: Does the AI Act – the European regulation that sets the rules for artificial intelligence – require that it be disclosed that a TV presenter is an AI-generated person?
The AI Act follows a risk-based approach. Certain practices are prohibited, and most compliance rules apply to (other) high-risk systems and general-purpose AI models such as LLMs. An AI-generated presenter does not fall within those categories.
However, Article 50 of the AI Act is relevant, as it sets out transparency obligations for AI systems that generate synthetic content. Incidentally, this article will not come into force until August 2026.
Article 50(1) stipulates that providers of AI systems intended to interact directly with people are obliged to inform those people that they are interacting with an AI system, unless this is abundantly clear from the circumstances. Although the fictional nature of the content will be anything but obvious to the young target audience, the key issue here is the requirement of interaction. Article 50(1) is typically relevant to chatbot-style applications. If the AI presenter presents the Top 20 with the communication going one way only and without any adaptation to individual viewer input, there will likely be no interaction within the meaning of Article 50 of the AI Act. However, if young viewers can ask questions and receive answers, the situation quickly changes.
Article 50(4) of the AI Act requires anyone deploying an AI system to generate image, audio or video content constituting a deep fake to disclose as much. However, the definition of a deep fake in the AI Act requires that the content bears a resemblance to an existing person and that the video content is falsely assumed to be authentic. If the presenter is entirely fictional – not based on an existing person – she falls outside that definition, strictly speaking. If it were up to the European Commission, the definition of a deep fake would be broadened significantly. The recently published Draft Guidelines propose a substantial lowering of the threshold for deep fakes, stating that content may constitute a deep fake so long as it resembles “someone or something that can exist or could have existed in reality. Under this definition, only unrealistic fantasy figures would be excluded. Evidently artistic works would then still fall under the attenuated disclosure obligation.
The AI presenter illustrates how the scope of Article 50 of the AI Act depends heavily on the specific technical and factual design. The AI Act does not impose a general disclosure obligation for all AI-generated media content, but sets out specific requirements that must be assessed on a case-by-case basis. As long as the AI presenter is not covered by the definition of deep fake and does not interact with the audience, the provider has no obligation under the AI Act to disclose that an AI presenter is being used.
Incidentally, under Article 50(2) of the AI Act, providers of AI systems used to generate the content (which need not necessarily be the party publishing the content) are obliged to ensure that the outputs of the AI system are marked in a machine-readable format and are identifiable as having been artificially generated or manipulated.
Moral obligation?
The announcement of the AI presenter has caused quite a stir, which is interesting in itself. It suggests that the public believes there is a moral obligation to be transparent about the use of AI in programmes aimed at children, regardless of the circumstances. However, the debate is fuelled in particular by the physical appearance of the presenter. After all, children are being presented with a picture-perfect girl.
Moreover, the AI girl is dressed rather provocatively, which is striking given that this is a programme aimed at children aged 6 to 12. With a human presenter, you could still say: “That’s just what she looks like.” But with an AI character, every feature is a conscious design choice. The hairstyle, the clothing, the physique, the proportions — these are all parameters set by humans. Some body parts even seem like ‘a fake within the fake’. For the creators, this is apparently part of an idealised image. Which does give rise to the question – especially when children are not aware that AI is involved –as to whether it is desirable to present such a fake character as reality.
The Kids Top 20 tv show is back. With a new presenter. One problem: she doesn’t exist. She has been entirely generated by artificial intelligence, looks perfect, is smartly dressed and has a physique that – even if she were a real person – does look artificial in places.
The accompanying NOS article raises the question of whether this is simply allowed. As a provider, are you required to disclose that your presenter is AI-generated? The answer is provided below, discussing the requirements set by the AI Act in this context.
The AI Act: a duty to disclose for an AI character?
Let’s start with the most obvious question: Does the AI Act – the European regulation that sets the rules for artificial intelligence – require that it be disclosed that a TV presenter is an AI-generated person?
The AI Act follows a risk-based approach. Certain practices are prohibited, and most compliance rules apply to (other) high-risk systems and general-purpose AI models such as LLMs. An AI-generated presenter does not fall within those categories.
However, Article 50 of the AI Act is relevant, as it sets out transparency obligations for AI systems that generate synthetic content. Incidentally, this article will not come into force until August 2026.
Article 50(1) stipulates that providers of AI systems intended to interact directly with people are obliged to inform those people that they are interacting with an AI system, unless this is abundantly clear from the circumstances. Although the fictional nature of the content will be anything but obvious to the young target audience, the key issue here is the requirement of interaction. Article 50(1) is typically relevant to chatbot-style applications. If the AI presenter presents the Top 20 with the communication going one way only and without any adaptation to individual viewer input, there will likely be no interaction within the meaning of Article 50 of the AI Act. However, if young viewers can ask questions and receive answers, the situation quickly changes.
Article 50(4) of the AI Act requires anyone deploying an AI system to generate image, audio or video content constituting a deep fake to disclose as much. However, the definition of a deep fake in the AI Act requires that the content bears a resemblance to an existing person and that the video content is falsely assumed to be authentic. If the presenter is entirely fictional – not based on an existing person – she falls outside that definition, strictly speaking. If it were up to the European Commission, the definition of a deep fake would be broadened significantly. The recently published Draft Guidelines propose a substantial lowering of the threshold for deep fakes, stating that content may constitute a deep fake so long as it resembles “someone or something that can exist or could have existed in reality. Under this definition, only unrealistic fantasy figures would be excluded. Evidently artistic works would then still fall under the attenuated disclosure obligation.
The AI presenter illustrates how the scope of Article 50 of the AI Act depends heavily on the specific technical and factual design. The AI Act does not impose a general disclosure obligation for all AI-generated media content, but sets out specific requirements that must be assessed on a case-by-case basis. As long as the AI presenter is not covered by the definition of deep fake and does not interact with the audience, the provider has no obligation under the AI Act to disclose that an AI presenter is being used.
Incidentally, under Article 50(2) of the AI Act, providers of AI systems used to generate the content (which need not necessarily be the party publishing the content) are obliged to ensure that the outputs of the AI system are marked in a machine-readable format and are identifiable as having been artificially generated or manipulated.
Moral obligation?
The announcement of the AI presenter has caused quite a stir, which is interesting in itself. It suggests that the public believes there is a moral obligation to be transparent about the use of AI in programmes aimed at children, regardless of the circumstances. However, the debate is fuelled in particular by the physical appearance of the presenter. After all, children are being presented with a picture-perfect girl.
Moreover, the AI girl is dressed rather provocatively, which is striking given that this is a programme aimed at children aged 6 to 12. With a human presenter, you could still say: “That’s just what she looks like.” But with an AI character, every feature is a conscious design choice. The hairstyle, the clothing, the physique, the proportions — these are all parameters set by humans. Some body parts even seem like ‘a fake within the fake’. For the creators, this is apparently part of an idealised image. Which does give rise to the question – especially when children are not aware that AI is involved –as to whether it is desirable to present such a fake character as reality.