AI omnibus: a necessary pause to enable real simplification

Europe’s ambition to lead in AI will only succeed if the AI Act can be implemented in a way that is credible, predictable and workable across sectors. The AI omnibus is therefore a pivotal moment not just to adjust timelines, but to determine whether simplification will be real.

At present, the proposal and surrounding debate miss the core issue. The need for a delay of the high-risk AI requirements is widely acknowledged: harmonised standards will not be ready, guidance remains pending and enforcement authorities are still being set up in several Member States. Without this infrastructure, compliance is not realistically possible.

But by bundling a necessary postponement with a modest set of substantive amendments, the AI omnibus proposal creates a strong political incentive for Council and Parliament to move quickly because the delay must enter into force in time. Delaying implementation and reforming substance require different timelines, different scrutiny and different political trade-offs. Treating them as one forces co-legislators to choose urgency over quality – and will lock in avoidable design flaws for years to come. That would be a huge missed opportunity.

That is why DIGITALEUROPE urges co-legislators to formally request a separate proposal postponing the entry into application of the high-risk AI requirements, to be adopted under accelerated procedures. This would provide immediate legal certainty for companies and authorities, avoid a compliance cliff driven by missing standards and guidance, and remove the artificial time pressure currently distorting the legislative debate.

Once the timeline risk is addressed, Council and Parliament can do what the omnibus was meant to enable in the first place: properly assess whether the AI Act works in practice, and where it needs to be improved.

Since early 2025, DIGITALEUROPE has been leading industry calls for AI Act simplification, which culminated with the publication of detailed recommendations in June 2025. This paper builds on these simplification recommendations, which reflect broad industry consensus, and responds to the Commission’s proposal by restating the key changes needed:

• Sectoral integration: The Commission’s own proposal to move medical devices to Annex I, Section B confirms what industry has long argued – AI requirements must be implemented through sectoral frameworks for regulated products. This logic should be extended to other comparable regimes, such as machinery and radio equipment.
• Legacy clause for existing systems: The legacy clause must be embedded in operative provisions, not left in recitals, to reflect how AI systems are developed, deployed and updated.
• Coherent data governance: AI-related processing of personal data should be governed through a single, GDPR-based framework, rather than parallel and prescriptive regimes that undermine accountability and flexibility.
• Proportionate obligations: Registration requirements, duplicative impact assessments, source code access provisions and CE marking for software should be removed where they add burden without improving safety.

None of these issues can be resolved properly under the current ‘adopt now or lose the delay’ dynamic. Separating the timeline adjustment from the substantive debate is therefore the condition for delivering an AI framework that is enforceable, investment friendly and fit for Europe’s industrial reality, prioritising implementation over new legislative initiatives or additional regulatory provisions.