Starting August 2, the European Union's AI Act requirements come into force. Companies must label deepfakes and other content created or altered by neural networks. This was reported by Qazaqyia.kz citing Kursiv Media.
The new rules apply to AI system developers and companies that professionally use such technologies. Specifically, providers of generative models must implement machine-readable labels that allow identifying the artificial origin of images, audio, video, or text.
Users must also be warned if they are interacting with a chatbot or other AI service.
Stricter requirements are set for deepfakes — realistic images, audio recordings, or videos that imitate the appearance, voice, or actions of real people.
Creators and distributors of such materials must clearly indicate that the content was generated or altered by artificial intelligence.
Similar requirements apply to AI-generated texts on socially significant topics if they are published without full editorial control.
For artistic, satirical, humorous, and fictional works, relaxed requirements may apply, but the artificial origin of the material must still be disclosed in an appropriate form.
The European Union fears that the development of generative technologies makes fake images, videos, and audio recordings increasingly realistic.
AI content can be used not only for creative purposes but also for spreading false information, political manipulation, and creating explicit materials involving real people without their consent.
One notable example is a fake video from 2022 in which Ukrainian President Volodymyr Zelensky allegedly called on soldiers to lay down their arms.
In 2024, explicit images of singer Taylor Swift created using AI without her consent also spread online.
Experts doubt that the new rules can be effectively applied in practice.
One problem is the lack of a unified technical standard. Companies use different labeling methods: watermarks, metadata, and file provenance tracking systems. These solutions are not always compatible with each other.
Moreover, digital labels can be lost during editing, compression, or transmission of images and videos across different platforms.
Another difficulty is that internet content is not limited by EU borders. A deepfake created outside the EU can quickly spread among European users.
The European Commission believes that one technology is not enough. For more reliable detection of AI content, companies are recommended to use multiple labeling and file provenance verification methods simultaneously.
Major AI developers, including OpenAI and Google, have supported European transparency requirements and joined a voluntary code of practice.
At the same time, industry representatives warn that content generation and detection technologies are changing rapidly, so regulatory requirements will have to be regularly updated.
The code of practice itself is voluntary. However, compliance with Article 50 of the EU AI Act is mandatory for companies operating in the European market.