BRUSSELS, BELGIUM / RankWire.AI / – European Union transparency rules for artificial intelligence took effect across the bloc on August 2, 2026. Article 50 of the EU AI Act now covers chatbots, deepfakes, synthetic media and some public-interest text. The law separates technical marking duties from notices that audiences can see. It does not require visible labels on every item created with AI. Each obligation depends on the system, the content and how an organisation presents it.
Providers of interactive AI systems must inform people when they are communicating with artificial intelligence. They may omit the notice when an informed user can clearly recognise the system’s artificial nature. Generative AI providers must also add machine-readable signals to synthetic text, audio, images and video. Those signals should help detection tools identify generated or altered material. Providers must use effective marking methods when available technology makes them technically feasible and reliable.
Organisations that publish synthetic content face a separate set of disclosure duties. They must identify deepfake images, recordings and videos that could appear genuine. They must also label AI-generated text about matters of public interest in certain cases. Human review and editorial control can remove that specific text requirement. A person or organisation must also accept responsibility for the published material before the exemption applies.
Visible notices target deceptive synthetic media
Article 50 also addresses emotion recognition and biometric categorisation systems that assess individuals. Operators must tell affected people when they use those tools, subject to limited legal exceptions. The law gives artistic, fictional, creative and satirical works more flexible disclosure treatment. Notices must not interfere with the normal display or enjoyment of those works. However, organisations still need an appropriate disclosure when such material includes deepfake content.
The European Commission has issued guidance explaining the transparency rules, definitions and available exceptions. It also introduced a voluntary code for developers and organisations that deploy generative AI. Participants may use the code to demonstrate how their marking and labelling methods meet the legal standard. Companies outside the code must establish other effective compliance measures. Optional icons can support public awareness, but the symbols alone do not prove compliance with the EU AI Act.
Authorities gain powers to enforce compliance
National market surveillance authorities carry primary responsibility for enforcement within EU member states. The European Data Protection Supervisor oversees relevant AI systems used by EU institutions and agencies. Regulators may impose fines reaching 15 million euros for violations of the transparency requirements. Penalties can also reach 3% of a company’s worldwide annual turnover. The regulation applies lower maximum amounts to smaller businesses when that calculation produces the lesser penalty.
A limited transition applies to generative AI systems that entered the market before August 2, 2026. Providers of those systems have until December 2, 2026, to meet the machine-readable marking duty. The extension applies only to that technical requirement. It does not delay notices for interactive AI, deepfake disclosures or rules covering public-interest text. The law also does not require retroactive labels for content produced before the transparency provisions took effect.