|
|
Since 2 August 2026 the transparency obligations of Article 50 apply: AI interaction and AI content must be recognisable, with exemptions for editorially reviewed content under human responsibility. Go through the 10 questions. Every "no" or "don't know" is a concrete work item. Orientation from implementation practice, not legal advice.
Marketing, sales, support, HR: AI texts and images are often created decentrally. Without an inventory there is no labelling strategy.
In AI interaction, people must be able to recognise that they are dealing with an AI, unless it is obvious. A clear notice in the chat window is usually enough.
Synthetic content must be marked as AI-generated in a machine-readable format (e.g. metadata/provenance standards). Check whether your tools deliver that.
An exemption applies to published texts if a human has reviewed them and carries editorial responsibility. Exactly this requires a defined approval or veto process, not just lived practice.
An approval log (who reviewed what and when) turns the exemption into solid proof, towards supervisors, customers, and in a dispute.
Fully automatically published AI content must be labelled. Your process should cover this case explicitly: automatic label instead of silent publication.
Whoever creates or distributes realistic image, audio, or video depictions of real persons or events must disclose their artificial origin. Also relevant for internal training or marketing purposes.
If external parties create content for you, they need your labelling and approval requirements, contractually and in practice.
AI literacy has been mandatory since February 2025. For Article 50 this concretely means: everyone who creates or approves AI content knows the labelling rules and their own process.
A lean AI policy that names tools, processes, and owners, and grows with new use cases. One document that is lived beats ten binders of theory.
Then a structured pass is worth it: AI inventory, risk classification, and implementing the missing items, usually done within a few weeks. We implemented Article 50 productively in our own platform Personativ (approval and veto principle, automatic labelling, proof log) and bring this hands-on experience into your organisation. All services at a glance: Implementing the EU AI Act.
A free intro call: we prioritise your open items and give an honest effort estimate.
Have further questions? Feel free to contact us directly via our contact form or by e-mail.
No. Published texts that a human has reviewed and for which someone carries editorial responsibility fall under an exemption. What matters is a defined approval process with proof. Fully automatically published content, however, must be labelled.
Synthetic images, audio, and video must be marked so that software can recognise the AI origin, for instance via metadata or provenance standards. Visible labels alone are not enough; the tool that creates the content must technically provide the marking.
The obligations apply to the actors along the chain, and as the client you carry responsibility towards your customers. Practically this means: labelling and approval requirements belong in the agency contract and in the shared working process. The legal assessment of individual cases is done by your law firm.