Amsterdam: The European Union has formally postponed the most demanding obligations of its Artificial Intelligence Act, after the Council gave final approval on 29 June 2026 to a simplification package known as the Digital Omnibus on AI. The move, which followed the European Parliament’s endorsement on 16 June, pushes the compliance deadlines for high-risk AI systems back by up to sixteen months and hands companies across the bloc a longer runway before the rules bite.
What actually changed
Under the amended timetable, obligations for stand-alone high-risk systems listed in Annex III now apply from 2 December 2027 rather than August 2026. High-risk AI embedded in regulated products under Annex I is deferred further, to 2 August 2028. The Commission argued that the original deadlines arrived before the harmonised technical standards and support tools companies need to comply were ready.
- Stand-alone high-risk systems: application deferred to 2 December 2027
- Embedded high-risk systems: application deferred to 2 August 2028
- A narrower definition of a “safety component” trims what counts as high risk
Guardrails, not just delays
The package is not a wholesale retreat. It introduces two fresh prohibitions, banning AI systems used to generate or manipulate non-consensual intimate imagery and child sexual abuse material. It also refines the definition of a safety component so that tools used purely for optimisation, convenience or quality control fall outside the high-risk net, unless a malfunction could endanger health or safety.
Industry relief, campaigner alarm
Technology firms and their trade bodies welcomed the breathing room, while digital-rights campaigners warned that repeated delays risk hollowing out the flagship law before it fully applies. The disagreement turns on whether predictability or momentum matters more for Europe’s credibility on artificial intelligence.
“Predictability matters as much as ambition. Companies cannot be asked to comply with standards that do not yet exist,” a Commission spokesperson said in defending the revised timeline.
Critics counter that every postponement weakens the signal that Europe intends to police artificial intelligence seriously, and that a statute amended before it takes full effect invites yet more lobbying.
Background
The AI Act, agreed in 2024, was billed as the world’s first comprehensive law governing artificial intelligence, sorting systems by risk and imposing the heaviest duties on uses that touch employment, credit, policing and critical infrastructure. The Digital Omnibus forms part of a wider simplification drive running through the Commission’s 2026 work programme, which pairs lighter administrative burdens with sharper enforcement in targeted areas. Readers can consult the Digital Omnibus on AI proposal for the full text.
What happens next
The amending regulation will be published in the Official Journal in the coming weeks and enters into force on the third day after publication, comfortably ahead of the next milestone in August 2026. National regulators must now issue guidance, while firms building high-risk systems gain time to document, test and register their tools before the deferred deadlines arrive.




