The new transparency rules of the AI Act come into force, requiring the labeling of AI content to distinguish real from fake and prevent abuse.
From Sunday August 2, many online content created using AI will have to be labeled within the European Union, via icons or labels, applying new transparency rules to help the public distinguish real from fake. Under the AI Act, the European Union requires companies to label their content AI, under penalty of a fine. “It is increasingly difficult for all of us to distinguish the real from the artificial,” explains a Commission official. “The challenge is to preserve citizens’ ability to trust what they see, hear and read.”
The most visible measure for the general public concerns the labeling of deepfakes, that is to say texts, images, videos or sounds appearing authentic, but which have been generated or manipulated by AI. Organizations or individuals who publish such content as part of a professional activity will now be required to report new content using AI. Violators risk fines.
A detailed system and precise obligations
Article 50 of the AI legislation applies from 2 August 2026. It sets transparency obligations for providers and deployers of certain AI systems, including generative and interactive AI systems and deepfakes. Public interaction devices using AI, such as chatbots, agents and other assistants, and virtual avatars, will need to be clearly presented as such. Suppliers of AI systems will also need to integrate markings and other digital watermarks to make it easy to detect content created using AI.
Texts intended to inform the public on topics of general interest, created using AI without human editorial control, will also have to be labeled. Brussels has published models of small icons (“AI”, “generated by AI”, “modified by AI”) which can be affixed to this content, and a guide has been put online to help companies comply with these obligations.
Existing AI systems will have until December 2 to adapt, and exemptions are provided, notably for “artistic, satirical, fictional or creative” works, or texts that are only edited with the help of AI.
Reactions and adaptation of digital players
The implementation of these new obligations provokes contrasting reactions among players in the sector. “We hear that it will be very complicated to implement. And that’s what we often hear about compliance requirements. And yet, the world continues to turn and we always end up finding solutions,” Ashley Casovan, who directs a center on AI governance within the Association of Data Processing Professionals, told AFP.
Several large platforms have already anticipated these rules by setting up their own labels. More than 3 billion pieces of content have already been labeled on TikTok, thanks to a set of tools available to creators and detection technologies, according to the platform. Meta has rolled out an “AI Info” label on Instagram and Facebook for posts using AI. Google, for its part, signed the European code of conduct on AI transparency, and emphasizes working with other companies including Nvidia, OpenAI and Apple on digital marking solutions.
However, some industry representatives warn of the potentially counterproductive effects of this regulation. “Instead of helping people, this risks sowing confusion,” warns Karen Massin, head of European affairs at Google. “If online content is flooded with AI labels and legal warnings, it will be harder for people to navigate.”
The AI-Generated Content Transparency Code of Practice was developed by independent experts through a multi-stakeholder process facilitated by the AI Office. The European Union thus intends to strengthen public confidence in digital information, while regulating the rapid development of artificial intelligence.
