The latest version of the Code of Practice on the labelling of deepfakes – what can the creative industry expect?

The AI Act[1] is the first comprehensive EU regulation governing the operation of artificial intelligence within the EU market, directly applicable in all Member States. Its aim is to ensure the safe and responsible development of AI whilst respecting fundamental rights and the values of the Union. One of the aspects that has been regulated is the obligation to label deepfake content.  To facilitate the understanding and application of these provisions at EU level, a Code of Conduct is being developed, the second version of which appeared some time ago on the European Commission’s website. What does this mean for those in the creative industries?

Deepfakes in the AI Act

Article 50 of the AI Act establishes transparency obligations for providers of AI systems and for entities that deploy and use them. This applies in particular to cases where AI systems are used to generate and manipulate images, sound or video material in a way that may give the impression of authenticity, i.e. so-called deepfakes. As a general rule, in such situations, entities using AI are required to clearly and explicitly disclose that the content in question has been artificially generated or modified using artificial intelligence.

However, this obligation is subject to certain limitations. The AI Act emphasises that obligations regarding the labelling of deepfakes must be implemented whilst respecting freedom of expression, as well as the freedom of art and science. Consequently, for content of an artistic, creative, satirical, fictional or similar nature, the transparency requirement is limited to disclosing the use of AI in a manner that is appropriate and does not undermine the reception of the work or its quality. Similar disclosure obligations apply to text content generated or modified by AI when published for the purpose of informing the public, unless it has undergone human verification or editorial review and is subject to editorial responsibility. This is what we know from the AI Act itself. Therefore, to facilitate understanding of Article 50 of the AI Act, the Code of Practice being developed is intended to serve as a practical tool to support providers and entities using AI tools in demonstrating compliance with the obligations arising from Article 50 of the AI Act.

First version of the Code

The first draft of the Code of Conduct on the transparency of AI-generated content was published in late 2025 as the result of extensive collaboration between representatives of industry, academia, civil society and EU Member States.

Following the publication of the first draft of the Code of Conduct, numerous voices emerged highlighting the need for a substantial revision. Expert analyses and research conducted by specialists from various fields showed that the original version was insufficiently transparent, overly complex and impractical for those required to apply it. In particular, concerns were raised regarding a lack of clarity regarding the scope of obligations and difficulties in translating general guidelines into concrete actions.

The second version of the Code

In response to the issues raised, a second version of the Code was developed, which is much better suited to practical needs. Simplifications were introduced and the structure of the document was streamlined, doing away with overly complex and difficult-to-apply classifications of AI content. In their place, clear and practical guidelines were proposed, particularly regarding content labelling, the use of icons and labels, and the fulfilment of transparency obligations.

Among the further measures set out in the Code, the development of a standardised icon to identify AI-generated content plays a key role. To this end, there are plans to establish a team of experts tasked with creating a standardised symbol applicable throughout the European Union. The proposed icon is intended to be interactive and to ensure consistency and recognisability of the labelling, regardless of the Member State or the type of entity using AI.

At present, the following icons have been proposed, possibly supplemented by an indication of the degree of AI use (e.g. ‘Generated with AI’, ‘Manipulated with AI’):

For audio content, the Code proposes a brief, audible disclaimer, sound or signal in simple and understandable language, disclosing the artificial origin of the audio. If the audio is of a longer duration, such information should also appear several times during the playback of the content. In the case of live-streamed deepfake video content, the disclaimer should be visible throughout the entire broadcast or, alternatively, displayed on screen at the start and at regular intervals during the broadcast.

An appropriate icon or disclaimer should be placed in a suitable and visible location, in accordance with the nature of the content and the context of its dissemination.

Artistic content

The Code of Practice, based on the exception set out in Article 50 of the AI Act, also provides for a specific approach to artists and creators of artistic works whose works contain elements generated or modified by AI. It is this provision that will most likely apply to the majority of entities in the creative sector. In such cases, the transparency obligation is limited to disclosing the fact that AI has been used. At the same time, the Code specifies that such disclosure should be visible at the latest upon the recipient’s first contact with the content, meet accessibility requirements, and be appropriate to the medium and context, so that the information regarding the use of AI is genuinely noticeable and not merely symbolic. The Code emphasises that the labels used must not, however, interfere with the reception of the work, its aesthetic qualities or creative freedom, nor hinder the normal enjoyment of the work. Information about the use of AI should be non-intrusive and adapted to the form of the work as well as the context of its presentation. At this stage, the Code proposes the following approach to labelling deepfakes:

• Real-time video: the label must be displayed no later than at the moment of the first viewing, in the top or bottom corner, for at least five seconds, with no further warnings during playback (e.g. during the opening credits);
• Video (non-real-time): the information shall be displayed for a sufficient duration to inform the viewer at the latest upon first viewing, without disrupting the viewing experience (e.g. during the opening credits);
• Image: the label must be displayed at the latest when the content is first shown in a suitable location, with the option of integrating it into the image or the image background, whilst ensuring the label remains recognisable to the viewer;
• Sound: a verbal disclaimer must be included no later than at the time of the first screening.

‘Contextual’ labelling is permitted if the above methods are not feasible or would affect the reception of the work. As an example, the Code cites placing an AI label outside the frame of a video or image, or audio content, but in close proximity to it, for instance where such content is displayed on websites or in apps. In the case of content presented in analogue form, e.g. at exhibitions, in art galleries or during festivals, information about deepfake content can be provided, for example, at the entrance, during ticket sales or as part of information provided in physical form.

Furthermore, in accordance with the Code, entities that regularly publish AI-generated content should have mechanisms in place to verify the accuracy of labelling. An important element of compliance is also ensuring appropriate training and awareness among staff involved in creating, modifying or publishing AI content.

What next?

The Code of Conduct treats deepfakes as a particularly sensitive category of content generated or modified by AI, due to their potential to mislead audiences and the risk of infringing personal rights and public trust.

The final version of the Code of Conduct is scheduled for May–June 2026, i.e. before the obligation to apply Article 50 of the AI Act comes into force. This provision will apply from 2 August 2026.

Source: https://digital-strategy.ec.europa.eu/en/library/commission-publishes-second-draft-code-practice-marking-and-labelling-ai-generated-content

[1] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 on the establishment of harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144, and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (the Artificial Intelligence Act).