Brussels, May 20, 2025 – The European Union has taken another significant step towards regulating artificial intelligence technologies and combating disinformation. Today, it became known that a preliminary agreement was reached between representatives of EU member states on key technical details and implementation mechanisms for the previously adopted landmark legislation – the AI Act. Particular attention during the current negotiations was paid to requirements for ensuring transparency and identification of content generated by AI systems, particularly so-called deepfakes.
The new agreements stipulate that developers and operators of powerful generative AI models capable of creating realistic but synthetic audio, photo, and video content will be obliged to implement reliable technical means for its labeling. These could be digital "watermarks," metadata, or other technologies that allow the end-user or a verification system to unequivocally determine that the content was created or significantly altered using AI. This measure aims to increase media literacy among the population and provide tools for recognizing potentially misleading information, which is especially relevant in the run-up to elections and in the context of hybrid threats.
Experts note that this decision is a compromise between the need to protect society from manipulation and the desire not to suppress innovation in the rapidly developing field of AI. Labeling requirements will be differentiated based on the risk level of the specific AI system. It is expected that the strictest rules will apply to systems capable of generating content that could be used for fraud, creating fake news, or discrediting public figures. At the same time, softer conditions may be envisaged for AI tools used for creative purposes with a clear indication of their nature.
The next stage will be the formal approval of these technical specifications and their subsequent implementation into the national legislation of EU countries. The technology community will need to develop and standardize effective labeling methods that are resistant to removal or bypass attempts. This EU decision is likely to have a significant impact on global standards for the development and use of generative AI, as many international companies strive to comply with high European requirements for access to the large EU market. It also stimulates discussion about the balance between innovation, freedom of speech, and the need for protection against malicious use of technology.