The EU AI Act has been in force since 1 August 2024, but many of its obligations are only gradually coming into full effect. Since 2 August 2026, this has been particularly true of Article 50, which deals with transparency and labelling requirements for AI content. As breaches of this provision fall under the AI Act’s sanctions regime, they can prove very costly for companies. Although the rule had been in the text of the legislation for some time, meaning companies knew what was coming, it is now a requirement to implement it.
1. Interactions with AI must remain recognisable. For example, in the case of an AI chatbot on a website – unless it is already obvious that no real person is responding.
2. Providers of generative AI systems must ensure that texts, images, videos and audio files generated or edited by AI are technically identifiable as such. Exceptions include, amongst other things, cases where content has been edited only slightly using AI and its meaning has not been substantially altered as a result.
3. Anyone using AI for emotion recognition or biometric categorisation must inform the individuals concerned and comply with the applicable data protection rules. Exceptions apply in particular to certain legally permitted applications in the field of law enforcement.
4. Organisations must disclose when they use AI to generate or alter deceptively realistic image, audio or video content. In the case of clearly artistic, creative, satirical or fictional content, appropriate labelling that does not unnecessarily disrupt the work is sufficient. AI-generated texts on topics of public interest must also, as a rule, be labelled unless the content has been sufficiently reviewed by a human or editorially checked, and a person or company assumes editorial responsibility for it.
5. Where labelling or information is required, it must be provided clearly and comprehensibly, and must be accessible to all.
6. The transparency obligations set out in Article 50 do not replace any other legal requirements. Companies must also comply with all other requirements under the AI Act, as well as other relevant EU and national regulations.
7. The European Commission, in collaboration with expert bodies, is developing practical guidelines designed to help companies correctly label and identify AI content. If these guidelines prove insufficient, the Commission may lay down binding common rules for implementation.
In practical terms, for film productions and businesses, this means that AI chatbots on websites must be clearly identifiable, as must AI-generated images, videos and voices, particularly if they are classified as deepfakes. A specific requirement for events is that if AI is used for emotion recognition or biometric categorisation, this must be clearly communicated.
This leaves two key questions: When is something a deepfake? And when does it constitute a matter of public interest?
The European Commission sets out three conditions that must all be met:
The content must bear a clear resemblance to the person, object, event or place depicted. It must relate to something real or plausibly possible. And it must be capable of giving viewers the false impression that it is genuine.
The third criterion in particular is a tricky one. Here, the Commission also takes into account the context, the target audience and the audience’s expectations. AI-generated backgrounds, special effects or traditional film post-production are therefore generally not considered deepfakes, provided it is clear to the audience that what is being shown is not an authentic, documentary-style representation.
This refers to socially relevant areas such as politics, security, health, the environment, fundamental rights or important economic and scientific developments – not every topic that is simply made public.
The European Commission provides three standardised labels for this purpose:
‘AI Basic’ identifies content created using AI, such as deepfake images, videos, audio or certain published texts, or where an additional note or further layer of information has been included.
‘AI Generated’ refers to content that has been generated entirely by AI, without human editing or editorial control beyond the mere prompt input.
‘AI Modified’ identifies content originally created by humans that has been altered using AI in such a way that it has become a deepfake or a text requiring labelling on a topic of public interest.
The icons are available for download on the European Commission’s website.
A breach of Article 50 may result in fines of up to 15 million euros or 3 per cent of the global annual turnover for the preceding financial year. For companies, the relevant maximum amount is generally applied; for SMEs, including start-ups, however, the more favourable rule applies, namely the lower of the two possible maximum amounts.
This could, for example, apply to:
a deepfake subject to labelling requirements that is published without disclosure,
an AI-generated text on a matter of public interest that is subject to labelling requirements,
the use of an emotion recognition system without informing the individuals concerned,
or an AI interaction in which, contrary to Article 50, users are not informed that they are communicating with AI.
No, but businesses should use AI more consciously in future. Furthermore, the fines mentioned are maximum limits: under the AI Act, sanctions must be effective, proportionate and dissuasive, meaning that the competent authorities will take the specific individual case into account – so not every minor error automatically results in the maximum penalty. Overall, the AI Act follows a risk-based approach: the greater the potential risk of an application, the stricter the requirements. For many everyday applications in communication, marketing and creative work, this does not automatically mean more red tape. At the same time, further requirements are being introduced, particularly for high-risk AI from the end of 2027/2028, and guidelines and technical standards are being continuously refined.
For businesses, this means above all you should be keeping an eye on developments, establishing clear internal processes and, when introducing new AI applications, checking at an early stage which rules apply. Those who use AI transparently and responsibly should therefore view the AI Act not so much as an obstacle, but rather as a framework for the safe use of a technology that has long been part of everyday life.