Speed vs Safety: CEN-CENELEC fast-tracks AI standards
The European Committee for Standardisation (CEN) and The European Committee for Electrotechnical Standardisation (CENELEC) have adopted exceptional measures to accelerate the development of key AI standards underpinning the EU AI Act. These measures include reducing the number of stages for publication and establishing a smaller drafting group to finalise the most delayed drafts. However, several members of the CEN-CENELEC technical committee have warned this could undermine consensus and lead to serious unintended consequences. While this move may provide clarity on the high risk requirements sooner, it creates uncertainty in the near term regarding the process, content and timing.
The balancing act
The AI Act relies on harmonised European standards, especially for high risk systems. Progress has lagged, prompting suggestions to delay parts of the Act, which is set to come into force in August 2026, while standards are still being developed.
The core tension is speed versus consensus. CEN-CENELEC is relying on existing rules to shorten procedure, while keeping public ‘Enquiry’ as a final step. They describe the measures as ‘exceptional’, ‘targeted and temporary’. Supporters say this balances urgency with transparency; critics say a smaller drafting circle and compressed steps weaken the ‘core principle of consensus’. The fast tracked drafts focus on elements of high-risk AI systems’ trustworthiness, datasets and bias. Drafters within CEN-CENELEC argue the streamlined process is likely to seriously harm standardisation. Timing remains tight: standards are due to be published by Q4 2026.
Comment
Clients should plan for dual track implementation of the AI Act, including through standards. Whilst adherence to available controls and current best practices is key, anticipate late cycle changes as fast tracked drafts move through Enquiry and undergo early revisions.
Policy risk is elevated. Calls to “stop the clock” may grow if standards slip, but there is no firm sign of a broad delay.
From a practical standpoint, clients should now focus on durable measures – such as data governance, risk management, transparency documentation and bias testing. This will ensure that any future adjustments based on changes to the standards are incremental.
In many fields, standards are used to facilitate interoperability of technology, such as in video encoding and telecommunications. In contrast, AI standards are proposed to be used to enable enterprises to meet AI Act requirements. As a result, there is a tension because complying with standards is usually optional, but complying with the AI Act is mandatory. Standards are typically agreed by consensus between many parties and yet in this case, consensus is proving difficult. This is particularly true for the trustworthiness, datasets, and bias of high-risk AI systems.
