2.2 The AIA and Its Requirements for "High-Risk AI Systems"
Regulating AI across various sectors in European society, the AIA is the first comprehensive regulation of its kind in the world.(1) Council of the EU, Artificial intelligence act: Council and Parliament strike a deal on the first rules for AI in the world (2 February 2024). The AIA currently applies within the EU member states.(2) The AIA Article 113. However, the EU has marked the Act as EEA-relevant, and it will likely be implemented in the EEA Agreement. Norway, for example, has already stated its intention to promptly incorporate the Act into Norwegian legislation following its implementation.(3) Government of Norway, Ministry of Local Government and Modernisation, Forordning om kunstig intelligens (KI-forordningen) (2025).
The AIA adopts a risk-based approach where an AI system is classified according to its level of risk. Some provisions in the AIA are general, while others are conditional upon the risk classification. AI systems can be deemed to pose minimal risk (not regulated), limited risk (certain requirements, primarily regarding transparency), "high-risk" (wide range of requirements), and unacceptable risk ("prohibited AI systems").
Both research questions of this thesis focus solely on the requirements for "high-risk AI systems". These are the requirements in Chapter III Sections 2-5 of the AIA. The focus on these specific requirements is due to their particular relevance for ANS, as illustrated in the introductory chapter of the thesis.
Pursuant to the AIA Article 113, the requirements apply from 2 August 2026. However, certain rules concerning notification come into effect on the 2 August 2025, and the obligations triggered by high-risk classification under the specific classification rule in Article 6(1) applies from 2 August 2027.(4) This is, however, impractical in the case of ANS. See Section 3.3.3 of the thesis.