Article 50 of Regulation (EU) 2024/1689, the EU AI Act, entered into application on 2 August 2026. The obligations are binding across all EU Member States without transposition. The date follows the staggered entry-into-force schedule under Article 113 of the Act, under which the majority of provisions apply from 2 August 2026 and the prohibitions on unacceptable-risk AI systems applied from 2 February 2025.
Article 50 imposes four categories of disclosure duty. Article 50(1) requires providers of AI systems intended to interact directly with humans to ensure those systems identify themselves as AI to users. Article 50(2) requires providers of AI systems generating synthetic audio, image, video or text to mark those outputs in a machine-readable format. Article 50(3) requires deployers using emotion recognition or biometric categorisation systems to inform natural persons exposed to that processing. Article 50(4) requires deployers of deepfake or AI-generated content on matters of public interest to disclose the artificial origin of that content. The European Commission published guidelines on 20 July 2026 to assist with implementation. The AI Omnibus provisional agreement of May 2026 provides that AI systems already on the market before 2 August 2026 that generate synthetic content have until 2 December 2026 to meet the machine-readable marking obligation under Article 50(2).
Providers of chatbots, voice assistants, and text-generation APIs serving EU users must now include AI identification disclosures in their user interfaces. Developers of text-to-image, text-to-video, and large language model applications must implement machine-readable watermarking or equivalent marking schemes, subject to the December 2026 grace period for pre-existing systems. Media publishers and advertising platforms that deploy AI-generated content addressing matters of public interest must label that content at the point of publication. Operators using emotion recognition systems in hiring or customer service, and those using biometric categorisation in their operations, must deliver prior disclosure to the individuals whose data is processed.
Article 2(3) excludes AI systems used exclusively for national security, defence, or law enforcement by or on behalf of a Member State authority from the Article 50 obligations. Content forming part of clearly artistic, creative, or satirical works is exempt from the deepfake labelling duty under Article 50(4). Providers of general-purpose AI models that supply foundation model outputs to downstream deployers may discharge their Article 50(2) marking obligations if they implement marking at model level and communicate the method to deployers who then comply at deployment level. The deadline for compliance with machine-readable marking for in-market systems under the AI Omnibus grace period is 2 December 2026.
Licentium advises clients on EU AI Act compliance and may assist directly or through its partner network. Providers, deployers, and enterprises affected by Article 50 obligations are welcome to contact us to discuss their compliance position. Work we undertake includes EU AI Act obligations assessments, transparency disclosure documentation, AI output marking implementation review, vendor due diligence for AI systems, and AI governance policy drafting.