'A text we can work with': Reaction to European Parliament vote on AI Act

Today’s European Parliament vote on the AI Act proposal marks a significant step towards setting a European AI rulebook and lays the foundation for good discussions with the Parliament, Council and Commission in trialogues. However, there remain some areas which can be improved to ensure Europe becomes a competitive hub for AI innovation. 

DIGITALEUROPE’s Director General Cecilia Bonefeld-Dahl said: 

• Whilst we still feel improvements can be made, this text is something we can work with and a good basis for further discussions.  
• On the positive side, the scope is a lot more precise, despite it being made wider. On general-purpose AI, it is right that deployers of high-risk uses should have most of the obligations. Self-assessment is welcome and will give companies much-needed flexibility. However, national authorities will need much more resources to hire AI experts and deal with the expected volume of paperwork.  
• Compliance was a key issue in our recent interviews with EU AI unicorns. AI innovators cannot afford lengthy delays nor legal uncertainty. Companies in every Member State need access to a sandbox to facilitate the risk classification process. They will also need a sufficiently long transition period to allow for the development of harmonised standards, with SME participation.  
• One area of concern is the two extra assessments that AI innovators will have to complete, which will create additional unnecessary burdens. Fundamental rights are already protected elsewhere in the Act, and sustainability requirements are not related to risk and are sufficiently covered in other EU legislation. 

Further reading 

DIGITALEUROPE’s recent sandboxing exercise tested the AI proposal on some of Europe’s most promising tech unicorns. See the results here. 

Six recommendations: 

1. There should be no regulation without a plan for investment, or the EU risks falling further behind other locations like the US and China. 
2. The proposal needs more clarity on scope. This has only become more important given late discussions on general-purpose systems, which might bring more companies into scope than originally envisaged in the Commission’s proposal.  
3. The existence of harmonised standards by the time the AI Act enters into application will be key to provide companies with clarity as to their exact obligations and implementation strategies.  
4. There should be a focus on the international level to align on standards and terminologies. For many participants the US market is key. The EU-US Trade and Technology Council is an excellent place to start.  
5. Gradual implementation and practical support will be key for SMEs. More funding and resources should come from the EU for compliance, including to make sandboxes easily accessible to a wide range of businesses.  
6. The AI Act should allow for modifications down the line – there should be a continuous sandboxing approach, focusing on adapting the text in the long run based on practical experience rather than purely aimed at compliance.