From Model Duties to Downstream Fundamental Rights: Contextual Accountability for GPAI Under the EU AI Act

Abstract

Under the EU Artificial Intelligence Act (AI Act), general-purpose AI (GPAI) raises a value chain problem for fundamental rights protection. GPAI-specific duties attach primarily to model providers, including requirements to produce and share certain model-level information, while downstream safeguards depend largely on the classification of the AI system and the context in which it is used. Yet rights impacts often become concrete only after a model is integrated into downstream systems and used in institutional settings. This paper examines when and why model-level GPAI obligations may fail to support downstream fundamental rights accountability. It develops a Contextual Accountability Gap Framework that distinguishes model-intrinsic, configuration-mediated, and use-context-dependent risks, and assesses whether upstream information remains specific, actionable, and contestable across the AI value chain. Through illustrative rights-sensitive scenarios, the paper shows how contextual accountability gaps emerge from the uneven translation of model-level information into system design, deployment oversight, and context-specific safeguards. It concludes by identifying three governance mechanisms that can better connect GPAI model duties to downstream rights protection: context-aware documentation, downstream translation records, and contextual fundamental rights review.