EU activates parts of AI Act

The European Union began applying another tranche of its AI Act on August 2, 2026, bringing into force transparency duties for certain AI systems, obligations tied to general-purpose AI, and enforcement of bans on prohibited practices. The European Commission said the new transparency rules require some AI-generated or manipulated content to be labeled and require users to be told when they are interacting with an AI system such as a chatbot or avatar. The law’s enforcement structure is split between the Commission’s AI Office, national market surveillance authorities and, for EU institutions, the European Data Protection Supervisor. At the same time, a separate EU amending law adopted in July delayed some of the bloc’s heaviest high-risk AI deadlines into 2027 and 2028. ### Which AI Act rules are already live? Article 50 transparency obligations started to apply on August 2, 2026, according to Commission guidance published on July 20. Those rules cover disclosure duties for systems that interact with people, as well as labeling duties for certain AI-generated or manipulated image, audio and video content, including deepfakes, and some text published on matters of public interest without human review or editorial control. (commission.europa.eu) The European Commission said August 2 also marked the application of rules aimed at “safer and more transparent AI,” with fines for breaches of the transparency rules reaching as high as 15 million euros or 3% of global annual turnover for companies, depending on the case. The Commission’s AI Act overview also says the law bans nine “unacceptable risk” practices, including harmful manipulation and deception, social scoring and some uses of real-time remote biometric identification for law enforcement in public spaces. (digital-strategy.ec.europa.eu) ### What did Brussels delay? Regulation (EU) 2026/1744, the “Digital Omnibus on AI,” was published in the Official Journal on July 24, 2026, after being adopted on July 8. The amending law changed the AI Act timetable for high-risk systems, pushing obligations for standalone Annex III high-risk AI systems to December 2, 2027, and for product-embedded Annex I systems to August 2, 2028. (commission.europa.eu) Those changes left the August 2, 2026 date in place for other parts of the AI Act. The result is a split calendar: transparency, prohibited-practice enforcement and parts of the general-purpose AI regime are already operative, while the main high-risk compliance regime now arrives later. ### Who is supposed to enforce the rules now? (eur-lex.europa.eu) The European Commission says enforcement is shared. The AI Office, which sits inside the Commission, has a direct role for general-purpose AI, while national competent authorities and market surveillance authorities supervise most AI systems at member-state level. The Commission says each member state should have designated and empowered national competent authorities by August 2, 2025. (digital-strategy.ec.europa.eu) A June 17, 2026 update from the independent EU AI Act implementation tracker at artificialintelligenceact.eu said the designation of national authorities remained a work in progress across Europe. That tracker says member states need market surveillance authorities, notifying authorities and, in some cases, fundamental-rights protection authorities, while general-purpose AI models are supervised centrally by the AI Office. (digital-strategy.ec.europa.eu) ### Why have there been no public cases yet? Forkast reported on September 1 that no public enforcement actions had been reported in the first 30 days after Article 50 became enforceable. Forkast said the quiet period reflected a “structural transition,” with regulators still building capacity and with enforcement for Article 50 resting mainly with national authorities rather than a single central enforcer. (artificialintelligenceact.eu) Forkast also reported that the AI Office had more than 125 staff across six units and two advisers and was running a recruitment drive for 40 contractual agents with applications due on September 8. That staffing push underscores that the EU-level machinery is still being built even as legal duties are already in force. ### What should companies watch next? (forkast.news) September 8 is the closing date for the AI Office recruitment drive cited by Forkast, and 2027 is the year the Commission says new EU evaluation capacity for AI models is expected to become operational. The next fixed AI Act milestones in the law itself are December 2, 2027, for Annex III high-risk systems and August 2, 2028, for Annex I product-embedded systems. (digital-strategy.ec.europa.eu) (forkast.news)