Article 50 of the EU AI Act applies in 38 days, and nothing in the AI Omnibus changes that. The transparency tier of Europe’s flagship AI regulation, politically the most visible piece because it governs what users see, holds firm for August 2, 2026, even as Brussels has just rewritten the calendar for almost everything around it.

The Omnibus, politically agreed on May 7 and formally approved by the European Parliament on June 16, pushed the Annex III high-risk regime to December 2, 2027, and the Annex I product-embedded regime to August 2, 2028. Article 50 wasn’t touched. The pattern is legible: the Commission will trade time on the hard-to-implement obligations, not on the ones the public can see.

Four obligations land together. Chatbot disclosure, machine-readable marking of generative outputs, disclosure for emotion-recognition and biometric categorisation, and labeling of deepfakes and AI-generated text. Only the watermarking duty under Article 50(2) carries grandfathering: systems already on the EU market before August 2 have until December 2, 2026 to comply. Chatbot disclosure has no grace period at all.

Fines reach €15 million or 3% of worldwide turnover. Latham & Watkins flags that where Article 50 conduct overlaps with the Act’s prohibited-practice provisions, exposure climbs to €35 million or 7%. The Omnibus also extends those prohibitions, from December 2, 2026, to non-consensual “nudifier” applications and AI-generated CSAM.

What’s still missing is the Commission’s final implementing guidance. The consultation draft closed June 3; the Code of Practice on marking and labelling AI-generated content followed on June 10. Sidley and Latham expect the final guidelines before the deadline they’re meant to interpret. Compliance teams are being asked to ship against a target the regulator hasn’t finished drawing.

Sources

Sources