Strategic postponement of parts of the AI Act: what changes and why
The European Commission is weighing a delay in the application of several blocks of the AI Act. We review which obligations may be postponed, which harmonised technical standards are at stake, the impact on the 2026-2027 timeline, and what companies should do to keep compliance on track.
An ambitious Regulation under pressure
EU Regulation 2024/1689 — the AI Act — is the world's most ambitious AI law. Adopted in 2024, its application was phased through 2027. During 2026, a political and technical debate has gained traction around the strategic postponement of some of its blocks.
Pressured by industry, Member States and growing global competition (United States, China, United Kingdom), the European Commission is weighing a slow-down of certain obligations without giving up the Regulation's backbone.
> Practical note: a postponement is not a repeal. Companies must keep their compliance roadmap; the risk of relaxing it is very high.
What is being considered for delay
The debate focuses on three blocks:
1. Harmonised technical standards
The AI Act refers to harmonised technical standards (developed by CEN-CENELEC) to specify requirements for risk management, data quality, cybersecurity and human oversight. Their drafting has lagged behind, and without them companies lack the "safe harbour" they need to presume conformity.
2. Effective sanctions regime
Although fine ceilings are set (up to €35 M or 7 % of global turnover), the operational deployment of national supervisory authorities and the AESIA — Spain's AI Supervision Agency — requires gradual rollout. Some Member States have requested a transition period without full sanctions.
3. High-risk system obligations embedded in regulated products
AI systems embedded in already regulated products (medical devices, toys, vehicles, machinery) have a longer schedule (2 August 2027). Proposals are emerging to align this calendar with the upcoming revisions of each sectoral framework.