Topic
Prohibited practices
Start with prohibited AI practices under Article 5 AI Act: every prohibition in plain terms, with examples, dates and fines.
The prohibited practices have applied since 2 February 2025 and carry the highest fine ceiling in the AI Act. They sound abstract but touch recognisable applications: emotion recognition at work, social scoring and certain forms of biometric categorisation. The line with permitted use is narrow, and that is exactly where the official guidance helps.
Official guidance on this topic
5- Guidelines on transparency obligations for providers and deployers of certain AI systems under Article 50 of the AI ActFinal guidelines (C(2026) 5054 final), adopted following consultation on the draft version of 8 May 2026. They clarify the scope, definitions and content of obligations for providers and deployers: designing systems so that people know they are communicating with AI, marking AI-generated or manipulated content in machine-readable form, informing people about deepfakes, about AI-generated text on matters of public concern without human editorial oversight, and about emotion recognition and biometric categorisation. Exceptions such as standard processing operations and artistic or satirical context are addressed. Separate paragraphs address the interplay with data protection law and refer to joint guidelines to be drawn up by the Commission and EDPB. Non-binding, but directive for how supervisors interpret Article 50. The obligations apply from 2 August 2026.PublishedGuidelinesEuropean Commission20 Jul 2026
- Commission Guidelines on prohibited artificial intelligence practices established by Regulation (EU) 2024/1689 (AI Act)For each prohibition, the Commission explains which AI practices are unacceptable and where the boundary lies, with legal explanation and practical examples: harmful manipulation, exploitation of vulnerabilities, social scoring, predictive policing based on profiling, indiscriminate scraping of facial images, emotion recognition at work and in education, biometric categorisation and real-time biometric identification at a distance. A separate chapter addresses the interplay with the GDPR, LED and EUDPR: by virtue of Article 2, paragraph 7, those remain unaffected, so a practice that just falls outside Article 5 can still be unlawful. The guidelines are not binding; only the Court of Justice provides binding interpretation. The formally adopted communication version bears the reference C(2025) 5052 final.PublishedGuidelinesEuropean Commission (DG CONNECT / AI Office)4 Feb 2025
- AI Regulation timeline (Algoritmekader)Dutch timeline from BZK showing which requirements apply on each effective date, linked to numbered requirements in the Algorithm Framework. Living document, last updated 3 August 2026, incorporating the postponement: new standalone high-risk systems from 2 December 2027, high-risk in products from 2 August 2028, with an exception until 2030 for systems already in use in government organisations on 2 December 2027. Note: the timeline does not list 2 August 2026 as the start date of the Article 50 transparency obligations; they have applied since that date.PublishedNational guidanceMinistry of the Interior and Kingdom Relations (Algorithm Framework)17 Jul 2025
- Supervision of AI takes shape: key role for the AP and the RDIResponse by the AP to the draft bill. The AP makes explicit which parts it will itself enforce: prohibited AI practices, transparency obligations such as the detectability of chatbots and deepfakes, and a large part of high-risk applications in work, education and government. The AP and RDI jointly provide coordination, knowledge-building and harmonisation, and will establish an AI regulatory sandbox from 2026 onwards. The AP stresses that parts of the Regulation already apply while the implementing legislation is not yet in force.PublishedNational guidanceDutch Data Protection Authority (AP), Algorithm Coordination Directorate20 Apr 2026
- Response of the President of the Administrative Jurisdiction Division to the draft AI Regulation Implementation ActLetter of 8 July 2026 from the President of the Administrative Jurisdiction Division to the Ministry of Justice and Security, with response and implementing assessment to the draft bill. The Division endorses that market supervision of courts by courts takes place, but advises that the sandbox task and the agreements on uniform interpretation of terms should not apply to its President because this conflicts with judicial independence. It advises expanding its own supervision to AI systems that fall only under the transparency obligations, and signals a question of competence when it is unclear whether a system is prohibited or high-risk. The implementing assessment estimates approximately 1 full-time equivalent additional.PublishedOpinionCouncil of State, Administrative Jurisdiction Division8 Jul 2026