HARNESS D3.3 Ethical Framework for Anticipatory Impacts - version 1

The EU’s Artificial Intelligence Act (AI Act), which entered into force in August 2024, establishes the world’s first comprehensive legal framework for AI governance, grounded in a risk-based approach and oriented toward protecting the fundamental rights of the EU Charter of Fundamental Rights. This deliverable critically examines the AI Act’s impact on fundamental rights, with particular focus on general-purpose AI (GPAI) models. In particular, it assesses how successfully the AI Act protects fundamental rights from the negative impacts of AI systems, with focus on GPAI. While the AI Act represents a landmark regulatory achievement, its current architecture contains structural limitations. Its risk-based approach presupposes that harms can be classified prior to deployment, which does not apply to GPAI models whose contexts and use are not yet known at the point of development. The Fundamental Rights Impact Assessment (FRIA) is largely confined to high-risk domains, leaving more diffuse or systemic impacts unaddressed. To address these gaps, this deliverable develops an ethical framework combining William D. Ross’ moral philosophy with Malgieri and Santos’ three-tiered approach to severity of rights-interferences. Making use of Ross’ concept of prima facie duties allows us to balance fundamental rights in the case of conflict. Malgieri and Santos’ approach enables us to assess the degree of impact on fundamental rights by an interference. As this is the first version of the deliverable, we introduce our work, explain the AI Act and its limitations, and present our ethical framework. The application of the introduced ethical framework and the anticipatory impacts assessment will follow in the second version.