Is an image or video genuine, or is it a fake created using artificial intelligence? Is a real person answering a telephone hotline, or is it a voicebot? Does the voice in an advert really belong to a celebrity, or was it generated by artificial intelligence?
Under new European Union transparency requirements that have applied since 2 August 2026, it must be clear in such cases that artificial intelligence has been used. The rules are part of the European Artificial Intelligence Regulation, known as the AI Act. They are intended to help people recognise artificially generated or altered content and protect themselves from deception, fraud and disinformation.
Users must be told when they are communicating with AI
Providers of chatbots, AI assistants or digital avatars must generally inform users from the outset that they are communicating with artificial intelligence rather than a person. An exception applies if this is already obvious.
Providers of generative AI must also ensure that artificially generated or altered texts, images, videos and audio files carry machine-readable labelling. This should enable technical systems to recognise such content as having been generated or altered using AI.
Deepfakes must be clearly labelled
Deepfakes must additionally carry a label that is clearly recognisable to people. Under EU law, the term refers to images, videos and audio recordings generated or altered using AI that depict existing people, places, objects or events so realistically that they could mistakenly be perceived as genuine.
The label must be visible or audible no later than when a person first encounters the content. A hidden label that can only be read by a technical system is not sufficient. Simplified requirements apply to clearly artistic, satirical or fictional works: the label may be designed so that it does not interfere with the presentation of the work.
Rules for texts created with AI
Texts generated or altered using AI must also be labelled if they are published without sufficient human review and are intended to inform the public about matters of public interest. These include, for example, politics, public administration, security, health, the environment, the economy or science.
No label is required if the content has actually been reviewed by a suitably qualified person and an individual or organisation assumes editorial responsibility for its publication. However, merely checking spelling and grammar does not count as sufficient human review.
Who do these obligations apply to?
The obligations mainly apply to providers of AI systems as well as companies, public authorities, organisations, associations, self-employed people and others who use AI professionally. Anyone who uses AI exclusively for private purposes and not in connection with professional or economic activity is generally not considered a professional operator of an AI system under the AI Act.
People must also be informed when emotion-recognition or biometric categorisation systems are used. Breaches of the transparency requirements can result in fines of up to €15 million. For companies, the fine may amount to up to three per cent of their total worldwide annual turnover in the preceding financial year.
Voluntary EU symbols for labelling
The European Commission provides special symbols that can be used to label content that has been created entirely by AI or partly altered using AI. Use of these symbols is voluntary. However, the legal obligation to clearly disclose certain content created or altered using AI applies regardless of whether these symbols are used.
Further information:
WDR: New EU rules on labelling AI content
Sources:
European Commission: Transparency obligations under Article 50 of the AI Act
European Commission: Guidelines on AI transparency obligations
Regulation (EU) 2024/1689 on Artificial Intelligence
tun26082603