Commentary. Europe’s experiment with artificial-intelligence regulation has reached an uncomfortable moment. The European Union has delayed key obligations for high-risk AI systems while simplifying implementation through its Digital Omnibus. The change may be sensible in parts. It is also a warning that ambitious rules can lose force when governments underestimate the practical burden of enforcing them.
The amended framework sets later deadlines: rules for stand-alone high-risk systems are scheduled for 2 December 2027, while systems embedded in regulated products are due to face obligations from 2 August 2028. The legislation also adds a prohibition on AI practices that generate non-consensual intimate or sexual content and child sexual-abuse material, according to the European Parliament’s legislative summary.
Our view is straightforward: implementation should be made workable, but delay must not become deregulation by exhaustion. If Europe wants its AI Act to mean more than a landmark text, it must use the extra time to build enforcement capacity, not merely to postpone difficult decisions.
The case for slowing down
There is a legitimate argument for the Omnibus. Companies cannot comply with rules they cannot interpret, and regulators cannot supervise systems without adequate technical expertise, testing standards and institutional coordination. The European Commission says the changes extend timelines and simplify administration, while preserving the broader framework’s risk-based approach.
That matters particularly for smaller firms. A compliance regime designed around large technology companies can unintentionally make market access more expensive for start-ups, universities and public-interest developers. Delaying obligations may give organisations time to document systems, recruit specialists and understand which uses qualify as high risk.
The EU’s approach is also more restrained than critics sometimes suggest. The AI Act does not prohibit artificial intelligence generally. It distinguishes between unacceptable practices, high-risk applications and lower-risk uses. That structure recognises a real difference between a system recommending a film and one influencing employment, education, access to services or other consequential decisions.
Why delay still carries a cost
Yet the social effects of automated decisions do not wait for a compliance deadline. AI systems are already used in workplaces and public-facing services, and their errors can affect income, opportunity and dignity. The EU’s own employment-focused guidance identifies recruitment and worker-management systems as high-risk areas and says affected workers should receive information before such systems are deployed.
Postponement therefore creates an accountability gap. Organisations may have commercial reasons to adopt systems now, while the strongest obligations arrive years later. Voluntary commitments can help, but they are not a substitute for enforceable rights when an applicant is rejected, a worker is monitored or a citizen cannot obtain a meaningful explanation.
The delay also complicates Europe’s claim to regulatory leadership. The AI Act entered into force in 2024 and was presented as the world’s first comprehensive AI framework. Leadership is not measured only by passing a law. It is measured by whether ordinary people can see its protections in practice.
The better bargain
The answer is neither maximal regulation nor a race to the lightest rules. Europe should pair the new timetable with visible interim safeguards. Regulators should publish clear guidance for high-impact deployments, require meaningful human review where automated decisions affect rights or livelihoods, and give workers and consumers accessible channels to challenge outcomes.
Governments should also report publicly on readiness: how many inspectors have been trained, how testing standards are progressing, how many complaints have been received and whether national authorities can investigate providers across borders. A delay without measurable milestones is simply an open-ended postponement.
Critics are right that poorly designed rules can suppress innovation. Industry is right to demand clarity rather than shifting interpretations. But the answer to uncertainty should be better public administration, not weaker protection. The EU should treat the next two years as a construction period for enforcement, with deadlines that are firm, transparent and difficult to move again.
Artificial intelligence will continue advancing while lawmakers revise schedules. That is precisely why accountability cannot remain an item marked “later.” Europe has bought time. It should spend that time proving that responsible innovation means protecting people before harm becomes a case study.