Most AI Act discussions revolve around high-risk systems. For many product teams, though, the transparency obligations under Article 50 are the ones that affect them first and most often, because they apply to ordinary chatbots and generative features — regardless of whether a system is classified as high-risk AI.
The four cases
First, providers must ensure that people interacting directly with an AI system are informed of this, unless it is obvious from the circumstances. Second, providers of systems generating synthetic audio, image, video or text content must mark it in a machine-readable format so that it is detectable as artificially generated. Third, deployers of emotion recognition or biometric categorisation systems must inform the people exposed. Fourth, deployers must disclose deepfakes they generate, and AI-generated text published to inform the public on matters of public interest, unless it has undergone human review with editorial responsibility.
What it means for product teams
A support chatbot needs a clear notice that an AI is answering, ideally at the start of the interaction and not in the small print. A feature generating images or text needs technical marking, for example via metadata or watermarks, as robust as the state of the art allows. For technical implementation the Commission published a code of practice on marking AI-generated content in June 2026 and adopted guidelines on Art. 50 in July 2026; teams should follow these rather than invent their own solutions.
Timing
Art. 50 has applied since 2 August 2026; the Digital Omnibus, Regulation (EU) 2026/1744, did not move that date. The only relief is a transition for machine-readable marking under Art. 50(2): generative systems placed on the market before 2 August 2026 must implement it by 2 December 2026; new systems must mark from day one. Content generated before 2 August 2026 need not be marked retroactively, but AI text published after that date must be. The chatbot notice and deployers’ disclosure duties apply without any transition.
Delimitation
Article 50 does not replace other obligations. A chatbot processing personal data remains subject to the GDPR or the Swiss FADP; a system that is also high-risk is additionally subject to high-risk obligations. And labelling does not protect against liability for wrong content: a correctly labelled hallucination is still wrong information.
Swiss providers
Swiss companies offering chatbots or generative features to users in the EU are covered as providers under the AI Act. Implementation is usually manageable: a visible notice, technical marking and a check of whether published AI texts are under editorial responsibility. We implement this with the product team as part of our AI Act consulting.