Is AI harder than cement? Is the EU AI Act being delayed?
Based on recent reports, the EU AI Act isn't being delayed, yet, but there are significant delays in the harmonised technical standards that companies rely on for compliance guidance.
What happened
“We need standards and guidelines to support the AI Act implementation before it starts to apply. We know that the AI Act already applies in part. We have to make sure that we are preparing everything in time for the next steps.”, EU digital chief Henna Virkkunen said this week. This comes hot on the heels of various claims that the EC is considering delaying the AI Act.
Why the standards delay matters
The standards were scheduled to be ready by August 2025 but will now be adopted in 2026, according to CEN-CENELEC, the European standardisation bodies responsible for developing them. The EU AI Act itself remains on track; the act will be fully in force by August 2026 for Annex III use cases and by August 2027 for Annex I.
However, these harmonised standards are crucial because they provide a "presumption of conformity" with the essential requirements of such legislation. In other words, if companies fulfil the Harmonised Standards, compliance with the requirement is presumed. Without them, companies must find alternative ways to demonstrate compliance, which is a more complex and costly process.
The European Commission may consider providing "alternative solutions" in case technical standards for companies seeking to demonstrate compliance with the EU's AI Act are not ready on time. However, it is unclear whether these alternative solutions will be acceptable to either the European Commission or industry.
Why the delay in standards is happening
Reaching a consensus on key topics has often proved challenging, and the process has been slower than anticipated, even by standardisation stakeholders. The delay stems from several factors:
Complexity: Given the scope of the various required standards, their complexity, the limited resources of the standardisation development organisations and an already tight timeline, the standards are unlikely to be adopted until 2026. However, some may be available for public enquiry (consultation) in 2025.
Consensus-building: The standards development process requires extensive consultation with and consensus-building among a diverse group of stakeholders. The responsible committee, CEN-CENELEC JTC 21, is one of the largest and most diverse European standards committees in memory.
Novel requirements: There are elements in the context of the EU AI Act that are not captured in any of the activities from existing international standards. The regulator is asking for standardisation in areas where there is no existing state of the art.
Has this happened before?
Yes, delays in harmonised standards are not unprecedented in EU legislation. Similar challenges have occurred with other complex regulatory frameworks:
Medical Device Regulations: There are ongoing delays in companies obtaining CE marks under the new regime, and given the delays in harmonisation of international standards under the EU MDR and IVDR, an approach to the use of 'state-of-the-art' standards was introduced.
The Radio Equipment Directive provides one of the clearest examples of how standards delays can cause compliance difficulties. Before EN 18031, no harmonised standards existed for certain cybersecurity obligations — forcing nearly all manufacturers to plan for expensive Notified Body involvement (rather than self-assessment) until these standards were finally harmonised in 2025.
The cement industry is particularly frustrated, as the revision process of the cement standard EN 197-1 has dragged on for over a decade. This forced the industry to pursue alternative approaches.
The Machinery Directive has also seen multiple waves of standard withdrawals and delays.
The AI Act delay is therefore part of a well-established pattern in which the complexity of modern EU regulatory frameworks consistently outpaces traditional standards development timelines, requiring alternative compliance pathways and interim solutions.
One might also imagine that AI is a little “harder” than cement.