The decision to exclude the European Telecommunications Standards Institute (ETSI) is part of a broader spat between the EU executive and the standardisation organisation, considered to be too driven by the private sector.
The European Commission published its draft standardisation request for the Artificial Intelligence Act earlier this week, a flagship proposal to regulate AI applications based on their capacity to cause harm.
The new draft of the request includes some significant changes compared to an early version that EURACTIV reported on in May. Most significantly, the Commission has excluded one of the three European standardisation organisations.
The request is now only addressed to the European Committee for Standardisation (CEN), and the European Committee for Electrotechnical Standardization (CENELEC), whilst ETSI has been downsized to a mere consulting role.
What sounds like a dry list of acronyms mostly known to tech geeks hides a political decision in the EU executive’s approach to standard-setting, an aspect that has gathered the increasing attention of world powers like the United States and China.
Industry standards will largely determine how the EU’s AI rulebook will be applied in practice, to the point that academics defined it as where ‘the real rulemaking’ in the upcoming regulation will occur.
European standard politics
In February, the European Commission set out its strategy to become more assertive in the way it participated in standard-setting, where it considered that non-European companies, particularly American and Chinese, have gained the upper hand.
The strategy came as a slap in the face to ETSI, which the Commission accused of being held hostage by non-European influences and requested internal reform to give more weight to national standardisation bodies. Such an internal restructuring has so far progressed slowly.
According to a Commission spokesperson, the decision was based on considerations related to the preparatory work of the standardisation bodies, the internal governance and the capacity to collaborate in the broader international context.
In particular, CEN and CENELEC established an AI-focused Joint Technical Committee in June last year and already have agreements in place with the International Organization for Standardization and International Electrotechnical Commission.
However, according to two sources informed on the matter, the question might also be personal, with ETSI’s director general Luis Jorge Romero, who has a great deal of leeway in managing the organisation, frequently clashing with EU officials.
Notably, Romero has been striving to assert independence from the EU executive, thanks to the fact that ETSI has managed to establish a successful business model, attracting tech companies from all over the world, whilst CEN and CENELEC still heavily rely on EU funding.
Relations were not improved after the Commission coordinated a response to prevent the renewal of an ETSI’s working group on IPv6+, a controversial technology conceived by Chinese telecom giant Huawei and aligned with the Chinese government’s view on internet governance.
Last week, ETSI’s general assembly elected a new chair, Bettina Funk, deemed closer to the Commission. One of the sources interpreted the move as an attempt for the organisation’s members to mediate with the EU executive, as Romero’s approach is increasingly deemed counterproductive.
However, although largely expected, the Commission’s decision shows that the appeasement move probably came too late or was deemed insufficient, as it seems unlikely it will be reversed.
ETSI declined to respond to EURACTIV’s request for comment.
Substantial changes
The content of the draft also includes other significant changes.
A reference has been added that CEN and CENELEC should consider the United Nations Convention on the Rights of Persons with Disabilities when developing the standards. Consumer organisations have also been included among the stakeholders to be consulted.
The European standardisation bodies will be required to be experts on fundamental rights. Still, civil society organisations have often lamented that technical standards are insufficient to protect fundamental rights.
In addition, the draft also includes alignment with internationally recognised terminology, a likely reference to the EU-US AI Roadmap that includes the commitment to develop a shared understanding around critical concepts such as risk and trustworthiness, with specific metrics to measure them.
A definition of state-of-the-art has been included, defined as a technical capability based on consolidated technology and scientific findings, leaving out experimental stage research and immature research.
For the first time, accuracy has been distinguished by statistical accuracy, addressing a technical error introduced in the draft AI Act.
CEN and CENELEC are also asked to consider costs for SMEs, particularly for the quality management system and conformity assessment.
The deadlines for all deliverables have been moved from 31 October 2024 to 31 January 2025, aligning them with the timeline of the legislative process.
[Edited by Nathalie Weatherald]
