Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into legal force today, three days after publication in the Official Journal on July 24 and six days before the transparency obligations it amends become enforceable across the EU’s 450-million-person single market. The compressed timetable is written into the law itself, an unusual admission that Brussels sequenced the rescue package too close to its own deadline.

Parliament adopted the text on June 16, the Council approved it on June 29, and it was signed July 8. What survives that legislative sprint is a bifurcated regime: transparency now, high-risk enforcement later.

From August 2, Article 50 requires chatbot disclosure, synthetic-content marking, and deepfake labelling. AI Office powers over general-purpose models activate the same day. Commission guidelines published July 20 extend disclosure duties to AI agents interacting with third parties, closing an ambiguity the agentic-AI industry had been quietly relying on.

The deferrals are the real story. Annex III high-risk systems, recruitment, credit scoring, law enforcement, border control, now have until December 2, 2027. Annex I products slide to August 2, 2028. Systems already on market before August 2 get grandfathered until December 2, 2026 for Article 50 compliance. The new Article 5 ban on AI-generated non-consensual intimate imagery, the “nudifier” prohibition, activates the same December date.

Latham & Watkins puts Article 50 fines at up to €15 million or 3% of worldwide turnover, and Article 5 breaches at €35 million or 7%. CDT Europe’s July bulletin reports MEPs are pressing the Commission to prioritise cyber-offense, biological, and loss-of-control risks. Anthropic’s Mythos model is already in Brussels’ sights.

The pattern is familiar from GDPR’s 2018 rollout: headline obligations arrive on time, the harder structural pieces get quietly rescheduled. Enforcement legibility beats enforcement readiness.

Sources

Sources