The European Union has formally agreed to postpone a large part of its AI Act. The so-called Digital Omnibus on AI, endorsed by the European Parliament on 16 June and given final approval by the Council on 29 June, defers the obligations for high-risk AI systems that were due to apply from 2 August 2026. Publication in the Official Journal is expected imminently.
The new calendar works like this. Obligations for stand-alone high-risk systems listed in Annex III of the AI Act (areas such as recruitment, credit scoring, education or access to essential services) move to 2 December 2027. Obligations for AI embedded in regulated products under Annex I (machinery, medical devices, vehicles) move to 2 August 2028.
What has not moved is the part many companies overlook. The transparency obligations of Article 50 still apply from 2 August 2026, a few days from now. In plain terms: if your chatbot or voice agent interacts with customers in the EU, it must be clear to them that they are dealing with an AI. And if you publish AI-generated or AI-manipulated content (synthetic audio, images or video, deepfakes included), it must be labelled as such, in a machine-readable way.
The rest of the framework also stays in place. The prohibitions on unacceptable practices have applied since February 2025, the obligations for general-purpose AI models since August 2025, and the AI literacy duty for staff continues, in a somewhat softened form. The omnibus even adds a new prohibition, effective 2 December 2026, against systems that generate non-consensual intimate imagery or child sexual abuse material.
What should a company do with all this? Three moves.
First, if you operate customer-facing AI anywhere in the EU, get the disclosure right now. Telling users they are talking to an AI is the cheapest compliance obligation you will ever meet, and from next week it is the law.
Second, if you produce synthetic media at any scale, build labelling into the pipeline instead of patching it on later. Retrofitting metadata across thousands of assets is exactly the kind of project nobody enjoys.
Third, use the deferral as it was intended: not to shelve the high-risk work, but to do it properly. Inventory your AI systems, classify them against Annex III, and ask your vendors the awkward questions about documentation and data governance while there is still room in the calendar.
The delay answered genuine industry concerns: technical standards were not ready and official guidance arrived late. But companies that read the omnibus as "regulation cancelled" will be the ones retrofitting compliance in a hurry through 2027. Transparency arrives now. Discipline pays either way.