The European Union has taken a decisive step in regulating artificial intelligence. Starting August 2, new rules will come into force in the region, obliging AI system developers to ensure transparency in interactions with users. The main focus of the legislation is on labeling generated content and identifying digital assistants.
According to documents published in the Official Journal of the EU, a strict requirement is introduced for chatbots and voice assistants: they must be designed so that the user clearly understands they are communicating with a machine, not a living human.
Technical requirements for labeling
Developers of generative systems are obliged to implement special technical markers into the results of algorithmic work. This applies to images, audio, video, and text. In particular, any publication of generated content containing images of real people must have a direct indication of its artificial origin.
Special attention is paid to informational materials. Texts created by AI to inform about socially important events are subject to labeling if they have not been verified by a human and a specific editorial board has not taken responsibility for them. At the same time, the requirements do not apply to systems performing only auxiliary functions, such as correcting spelling or grammar.
It is important to note that fantastical content, for example, images of people flying in the air, does not fall under the definition of deepfakes.
What will the average user see?
Despite the rules coming into force, mass visual changes on user screens may not be immediately noticeable. According to Article 50, paragraph 2, priority is given to machine-readable labeling. Developers are implementing metadata or hidden watermarks that are read by special analysis tools, not by the human eye.
The use of visual icons by the European Commission (to denote fully generated, partially modified, or AI-assisted content) remains voluntary within the framework of the Code of Practice.
Transition period and sanctions
For systems already placed on the market before August 2, 2026, a transition period is provided. Developers have an additional four months — until December 2, 2026 — to implement the necessary labeling mechanisms.
The AI Act provides for strict financial consequences for violating transparency requirements. Fines can reach 15 million euros or amount to 3% of the company's global annual turnover.
From December 2, 2026, a full ban on offering AI systems that create intimate images without the consent of the persons depicted or materials depicting sexual violence against children will come into force in the EU. Violation of this ban is subject to maximum sanctions — up to 35 million euros or 7% of annual turnover.
Stricter norms regarding high-risk systems (high-risk AI) will be introduced in stages later: in December 2027 and August 2028.
