Ten steps to make the AI Act an EU success story

The AI Act is a complex and rather complicated piece of legislation that risks hampering the competitiveness of the European AI ecosystem. Since AI is the key driver of the current digital transformation processes, the gap between EU and other regions could even be widening as a result. Therefore, it seems to be key to simplify the compliance with the AI Act for EU providers and deployers, in particular by avoiding unnecessary bureaucracy, clarifying legal uncertainties and better supporting innovation in AI. For me, the following steps are required to make the EU AI Act a genuine success story:

1. Harmonized technical standards: The Commission shall issue a public statement (i.e. a communication or a paragraph in the AI Pact) that the work on horizontal and vertical standards in CEN/CENELEC, including those on GPAI models, needs to urgently accelerate and that European companies are encouraged to extensively engage in the process. The use of harmonized technical standards must be the key compliance tool for European SMEs, including start-ups, as this would massively reduce the compliance costs with regard to the AI Act.
2. Legal certainty with harmonized guidelines, model contractual terms and templates: In order to address and specify certain unclear or even contradicting elements of the AI Act, the Commission shall fulfil its obligation to provide the guidelines for Article 6(2c) as soon as possible but at least six months before the respective parts of the AI Act become applicable. The same deadline shall apply for the guidelines mentioned in Article 82a, the QMS guideline (Recital 74a / Art 55a(2b)), the model contractual terms in Article 28(2b), the FRIA template (Recital 58g / Art 29a(5)), the copyright template (Recital 60k, Art 52(1d)), and the SME templates (Recital 73, Art 55(2a(a))). By providing this kind of secondary legislation, the Commission will ensure that there is one common interpretation of the AI Act within the Single Market, making sure that the law is implemented in an innovation-friendly way.
3. Legal overlaps: The Commission shall perform an in-depth gap analysis and identify contradictions and overlaps between the AI Act and other horizontal or sectorial laws. It should then issue a communication in which it clarifies the relationship between those laws (i.e. lex specialis, lex generalis, and complementary laws). The expert group together with Member States, as explained by the Commission in its Coreper statement, shall play a leading role during this exercise.
4. Governance overlaps: The Commission shall perform an in-depth assessment of the relationship between the AI Act governance bodies and the existing governance bodies on EU and Member State level, including sectorial governance bodies. It should then issue a communication in which it clarifies which governance body is the lead authority and how the different bodies should cooperate with each other.
5. Concept for sandboxes: The Commission shall immediately start drafting implementing acts in accordance with Article 53a with the aim to provide common principles on the application of the participation in a Regulatory Sandbox. They shall also, based on a comprehensive strategic concept, encourage the creation of industry- and sector-specific Regulatory Sandboxes instead of having multiple but very similar national Regulatory Sandboxes that are competing with each other.
6. Easy compliance for SMEs: Smaller market players within the EU cannot afford high compliance costs, especially if we want them to be globally competitive. The Commission shall therefore develop, in close cooperation with the relevant stakeholders, a tool that allows AI companies to check their compliance with the AI Act themselves without drawing on expensive third party auditing or certification. The results of the compliance tool shall give SMEs, including start-ups, a presumption of conformity.
7. Training data and the access to high quality datasets: The AI Act does not address the two intertwined key problems of EU AI developers of (a) how to train their AI systems without violating data protection, copyright or personality rights and (b) how to access high-quality datasets. The Commission shall provide specific guidelines addressing both these obstacles for EU AI developers, in particular by summarizing the relevant rules in different EU laws (i.e. AI Act, GDPR, Data Act, DGA, Copyright Directive). This summary shall be complemented by harmonized procedures (i.e. point of contacts, deadlines), common interpretations (i.e. how to anonymize, pseudonymise or use synthetic data), and a new understanding of Art 6(4) GDPR with regard to the training of algorithms and their datasets (i.e. rethinking purpose limitation; reducing the focus on ‘consent’ and the obligation to inform; measures to minimize red tape).
8. Concrete strategy on ‚Innovation in AI‘: The recent communication on ‘AI start-up and innovation’ is rather vague and misses the chance to connect the few innovation aspects of the AI Act (i.e. Regulatory Sandboxes, real world testing) with concrete practical measures. Moreover, it does not explain, which role the new EU governance structure (i.e. AI Office, AI Board) can play in promoting innovation in AI. Therefore, the Commission shall present a comprehensive AI roadmap for the EU with concrete investment sums, strategies to attract more venture capital, and covering topics that did not play a role so far in the regulatory debates on AI (i.e. leveraging synergies with the green deal, military use of AI), taking note of the AIDA roadmap of the European Parliament. The upcoming political discussions on the Multiannual Financial Framework (MFF) shall take into account the new expenses for the AI Office, Regulatory Sandboxes, and the HPC capacities. In particular, the budget of the Digital Europe program must be raised significantly.
9. Attract talent for the AI Office: The governance of the AI Act will only work with a sufficiently staffed AI Office. For the 80 external positions, the Commission shall offer adequate salaries in order to attract genuine AI experts and thereby guarantee a balanced mix of nationalities, backgrounds and expertise. For the 20 internal places, staffers from DG Competition shall be
   included in a significant number. Moreover, the AI Office shall be closely cooperating with the DSA/DMA enforcement unit.
10. Penalties: The Commission shall suspend and shall encourage Member States to suspend the issuing of fines in accordance with Article 71 on EU and on Member State level until all previous nine steps have been fulfilled and a sufficient level of legal certainty of the AI Act obligations for providers and deployers has been established.

https://www.axel-voss-europa.de/wp-content/uploads/2024/03/AVoss_TenSteps_AIAct-1.pdf