EU AI Act: Robustness & Cybersecurity

The EU Artificial Intelligence Act (AIA) defines varied legal principles for different AI systems. While some work has clarified aspects of these principles, robustness and cybersecurity have received limited attention. This paper addresses this gap.

Legal Challenges

We identify legal challenges and shortcomings in provisions regarding robustness and cybersecurity for both high-risk AI systems (Art. 15 AIA) and general-purpose AI models (Art. 55 AIA). Robustness and cybersecurity require resilience against performance disruptions.

Implementation Challenges

We assess potential challenges in implementing these provisions, considering recent advancements in machine learning (ML). Our analysis informs the development of harmonized standards, European Commission guidelines, benchmarks, and measurement methodologies under Art. 15(2) AIA. This aims to better align legal terminology and ML research, enhancing research and implementation efforts.