The EU’s AI Act is about to turn synthetic media disclosure from a policy discussion into a production and distribution checklist.
From 2 August 2026, Article 50 transparency obligations apply to certain AI systems used in the EU. The rules cover several different situations: telling people when they are interacting with an AI system, marking AI-generated or manipulated outputs in machine-readable ways, and labelling deepfakes and certain AI-generated or manipulated text published to inform the public on matters of public interest.
For media companies, the useful point is not that every AI-assisted edit suddenly needs a flashing warning sign. It is that teams will need to know which assets were generated or materially manipulated by AI, where those assets travel, and whether a disclosure or label is required before publication or distribution in Europe.
The European Commission has also published a Code of Practice on transparency of AI-generated content. It is voluntary, but it is designed to give providers and deployers a recognised way to show how they are meeting the AI Act’s marking and labelling requirements. Companies that do not sign the code are not automatically in breach, but they will still need to demonstrate compliance through other adequate measures.
That distinction matters. The code is not a magic compliance sticker. It is closer to a common operating manual for marking, detection and labelling practices, subject to assessment by the Commission and the AI Board.
The obligations are likely to touch several ordinary media workflows: AI-generated marketing clips, synthetic presenter or performer material, chatbot interfaces for viewers or subscribers, AI-generated public-interest explainers, localization assets using synthetic voices, and distribution packages that move across EU platforms or partners.
The awkward part is that many content pipelines were not built to preserve this information cleanly. A synthetic image may pass through creative review, retouching, versioning, transcoding, localization and platform delivery before anyone asks whether the final asset still carries the right metadata or visible disclosure. Compliance will depend less on one clever watermark than on boring but necessary handoffs between legal, editorial, post, marketing and media operations.
Companies preparing for the August deadline should map where generative AI enters their workflow, decide who is responsible for labelling decisions, document exceptions, and check whether suppliers can pass through the marking or provenance information they promise. If a vendor’s answer is mostly vibes and a dashboard, that is not a compliance programme. It is a sales demo wearing a tie.