The Missing Layer in EU AI Governance: Design-Level Risk, Structural Instability, and Policy Recommendations
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This white paper identifies a structural gap in current AI governance frameworks, using the European Union’s EU AI Act as a primary reference point. While existing approaches are largely built around risk classification, compliance obligations, and application context, this paper argues that such frameworks do not fully account for how AI systems behave in practice particularly as they scale, iterate, and integrate across environments. The central argument is that AI risk is not a static attribute tied to predefined categories, but a dynamic phenomenon that emerges along a continuous causal chain: design → system evolution → application Within this chain, a critical intermediate layer remains insufficiently addressed: design-level risk. This paper defines design-level risk as the set of structural conditions that shape how AI systems behave over time, including architectural decisions, coupling dynamics, stability under scaling, and the degradation of controllability as capabilities expand. The absence of this layer in current governance models leads to a structural misalignment between regulatory assumptions and system reality. As a result, implementation challenges such as delays in standardization, evolving guidance requirements, and misaligned timelines are not merely operational, but structural in origin. To address this gap, the paper proposes a set of policy extensions to existing frameworks, including: the introduction of design-level risk as a complementary governance dimension the integration of structural stability assessment into high-risk evaluation the establishment of continuous monitoring obligations for evolving systems the alignment of standards with observed system behavior the development of regulatory feedback loops linking deployment and governance These proposals do not replace existing regulatory approaches. They extend them enabling governance to engage not only with how AI is used, but with how AI systems evolve. The paper further suggests that the challenges observed in the EU context are not isolated, but indicative of a broader limitation in current AI governance paradigms. As AI systems continue to evolve, effective governance will depend not only on classification and compliance, but on the ability to understand and regulate system dynamics at the level where risk is formed. Keywords EU AI Act, AI governance, artificial intelligence regulation, general-purpose AI, AI risk, structural risk, design-level risk, AI policy, system stability, AI safety
Fields of Study & Subjects
Digitalization
Law
and Regulation
Ethics and Social Impacts of AI
Safety Systems Engineering in Autonomy
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 28 March 2026
- Language
- EN