BRUSSELS, BELGIUM / RankWire.AI / – New European Union transparency rules for artificial intelligence took effect on August 2, 2026. Article 50 of the EU AI Act now governs how companies identify certain AI interactions and synthetic content. The law covers chatbots, deepfakes, generated media and some public-interest text. It does not impose visible labels on all material produced with artificial intelligence. Instead, each duty depends on the system, the output and how an organisation presents it.
Companies operating interactive AI systems must tell people when they are communicating with artificial intelligence. The rule does not apply when an average user can clearly recognise the system’s artificial nature. Generative AI providers must also embed machine-readable information in synthetic text, images, audio and video. That information should support automated detection of generated or altered material. Providers must apply the measure when available technology can deliver reliable and effective marking.
Organisations that distribute synthetic media face separate public disclosure rules. They must identify deepfake images, audio or video that could appear genuine. They must also label AI-generated text about matters of public interest in certain circumstances. The text requirement does not apply when people review the material and exercise editorial control. A named person or organisation must also accept responsibility for publication before that exemption applies.
Public disclosures focus on deepfakes and civic information
The rules also apply to emotion recognition and biometric categorisation systems that assess individuals. Operators must inform affected people when they use those technologies, except in limited cases authorised by law. The EU AI Act gives creative, fictional, artistic and satirical works more flexible disclosure treatment. Notices must preserve the normal presentation and enjoyment of those works. However, operators must still provide an appropriate disclosure when a work contains deepfake material.
The European Commission released guidance to explain the scope of Article 50 and its exceptions. It also published a voluntary code for companies developing or deploying generative AI systems. Participants can use the code to show how their labelling and marking processes meet the legal standard. Companies that do not join must rely on other effective compliance measures. Optional icons can support public awareness, but those symbols do not establish compliance by themselves.
Regulators can impose significant financial penalties
National market surveillance authorities will lead enforcement across EU member states. The AI Office will supervise a narrower group of systems connected to general-purpose AI models and major platforms. The European Data Protection Supervisor will oversee relevant systems used by EU institutions and agencies. Regulators can impose fines of up to 15 million euros for breaches. The law also allows penalties reaching 3% of a company’s worldwide annual turnover.
Generative AI systems already on the market before August 2 receive a limited transition period. Their providers have until December 2, 2026, to meet the machine-readable marking duty. That extension applies only to the technical requirement for detectable synthetic content. It does not delay chatbot notices, deepfake disclosures or rules covering public-interest text. The law also does not require organisations to label content created before the transparency provisions took effect.