Article 50 of Regulation (EU) 2024/1689 (the EU AI Act) entered into force across the EEA on 2 August 2026. No national transposition is required; the obligations apply directly. The provision targets four categories of AI deployment: systems that interact directly with natural persons, including chatbots and voice assistants; emotion-recognition and biometric-categorisation systems applied to individuals; AI systems that generate or manipulate synthetic audio, images, or video; and AI systems that generate or manipulate text published for informational purposes. The obligations apply regardless of whether the system carries a high-risk classification under Title III of the Act.
Article 50(1) requires providers of AI systems designed to interact directly with natural persons to ensure those systems notify users they are communicating with AI, unless this is obvious from context. Article 50(2) requires deployers of emotion-recognition or biometric-categorisation systems to inform every individual subject to such processing. Articles 50(3) and 50(4) require providers and deployers of synthetic-content AI systems to mark outputs in machine-readable format and to label content as AI-generated or AI-manipulated for recipients. The European Commission adopted final guidelines on 20 July 2026 covering chatbot disclosure implementation, synthetic-content watermarking, and duties for general-purpose AI model providers.
Chatbot operators, synthetic-media platforms, AI customer service providers, voice assistant operators, and general-purpose AI model providers face the most immediate compliance requirements. Deployers that apply emotion-recognition tools in employment screening, educational assessment, healthcare triage, or access-control settings must issue Article 50(2) notices to every individual before the system operates. National market surveillance authorities designated under Article 74 enforce Article 50; non-compliance can result in fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher.
A transitional provision applies to the machine-readable marking and detection obligations in Articles 50(3) and 50(4). Providers of in-scope synthetic-content systems already placed on the EEA market before 2 August 2026 have until 2 December 2026 to implement compliant marking. Article 50(6) permits member states to displace the disclosure obligation where it would compromise criminal investigations or conflict with specific national law. Open questions remain about the technical specifications for machine-readable watermarking of AI-generated text.
We may advise on EU AI Act compliance and can introduce clients to specialists in our partner network. Contact us to discuss your organisation's obligations under Article 50. Work we undertake includes AI Act compliance assessments, Article 50 disclosure audits, synthetic-content marking reviews, and AI governance strategy across EEA jurisdictions.