BRUSSELS, BELGIUM / RankWire.AI / – European Union has started implementing new requirements for transparency concerning some AI-generated materials, effective across the member states as of August 2, 2026. Under Article 50 of the EU AI Act, obligations are now imposed on both creators and users of artificial intelligence systems. These regulations pertain to chatbots, synthetic media, deepfakes, and certain publicly relevant texts. Importantly, they do not mandate a visible warning on every AI-produced item. Instead, the law stipulates different compliance standards depending on the type of system, content, and publication method.
Any AI system that interacts directly with individuals must indicate that the user is engaging with artificial intelligence. This obligation does not apply if the system’s nature is clear to an attentive and knowledgeable observer. Developers of generative AI are also required to embed machine-readable markers into synthetic texts, audio, images, and videos. These markers aim to assist detection tools in identifying artificially generated or manipulated content. The regulation applies when the technology allows for reliable marking without impairing the output’s usability or altering its normal functioning.
Separate disclosure responsibilities are assigned to organizations publishing or presenting synthetic content. They must identify deepfake images, recordings, and videos that may seem authentic, as well as label AI-generated texts that address public-interest issues. However, this requirement does not apply after a significant human review and editorial oversight. The publisher or organization must also accept responsibility for the publication. All notices must remain clear, accessible, and visible during initial interactions or exposures.
Deepfake content faces mandatory disclosure rules
The scope of Article 50 also includes tools used for emotion recognition and biometric categorization involving individuals. Operators are required to inform people whenever these systems analyze them, except in limited legally defined cases. Works of a creative, artistic, satirical, or fictional nature are subject to less strict disclosure requirements. Notices for such content must not interfere with their usual presentation or enjoyment. Nonetheless, the obligation to disclose deepfakes remains in place in a suitable manner. Certain law-enforcement applications may qualify for exemptions if authorized by law and accompanied by safeguards.
The European Commission has issued guidance and a voluntary code designed to promote uniform compliance among member states. Providers and users can adopt the code to demonstrate how their marking and labeling systems align with legal standards. While participation is voluntary, the transparency requirements are compulsory. Companies outside the code are expected to implement other effective measures and provide explanations to regulators. Although optional icons may assist audiences in recognizing synthetic material, using only an icon does not guarantee full compliance with the AI Act.
Enforcement led by national regulators
Primary enforcement of these transparency rules rests with national market surveillance authorities. The European Data Protection Supervisor oversees systems used by EU institutions and agencies, while the AI Office supervises specific cases involving general-purpose AI models and major online platforms. Violations can result in fines up to 15 million euros or 3% of global annual turnover. Smaller companies face lower maximum penalties as dictated by the regulation. Member states are also empowered to issue warnings and other corrective actions through their national enforcement mechanisms.
A transitional period is in place for generative AI systems introduced before August 2, 2026. These providers have until December 2, 2026, to comply with the machine-readable marking requirement. This extension only applies to that specific technical obligation and does not delay the implementation of chatbot notices, deepfake disclosures, or regulations concerning public-interest texts. Content created prior to August 2 does not need retroactive labeling, but all new content and interactions from that date onward are subject to the transparency rules.