For months, much of the market has repeated that "the whole AI Act becomes applicable on 2 August 2026". That was true under the original text. It no longer is: the European Parliament voted the Digital Omnibus on AI on 16 June 2026, the Council adopted it on 29 June, and it entered into force in July. Here is precisely what changes, and what does not.

What the Omnibus changes

  • Annex III high-risk systems (HR and recruitment, credit scoring, biometrics, education, essential services): obligations deferred to 2 December 2027.
  • Annex I high-risk systems (AI embedded in regulated products: medical devices, machinery, toys): deferred to 2 August 2028.
  • Transparency for AI-generated content (Article 50): labeling becomes enforceable on 2 December 2026. For marketing and communications teams, this is now the nearest deadline.

The reason for the deferral is explicit: harmonised standards and technical guidance are not ready. The extra time is meant to let standardisation catch up, not to soften the substance of the obligations, which remain unchanged.

What does NOT change

  • Prohibited practices (Article 5) and the AI literacy obligation (Article 4) have applied since 2 February 2025. If your teams use Copilot, ChatGPT or AI-powered HR tools without documented training, the gap is already actionable.
  • General-purpose AI (GPAI) obligations have applied since 2 August 2025: technical documentation, copyright policy, training-data summaries.
  • Governance and penalties are in place, and national supervisors are operational, including Spain's AESIA, the first dedicated AI agency in the EU.

What to do with the extra time

The deferral is not an amnesty: it is time to prepare methodically instead of facing an audit in a rush.

  • Now: inventory the AI systems actually in use (including shadow IT), classify them by risk level, adopt an internal AI policy, and document Article 4 training.
  • Before 2 December 2026: set up AI-content labeling (internal taxonomy, visible labels, metadata) integrated into your tools and editorial guidelines.
  • 2027: technical documentation and human oversight for systems classified as high-risk, with an evidence file ready for the first inspection.

Companies using AI today already have obligations today. The right question is not "when will this affect us", but "which of these obligations affect us already".