Trustworthy AI Through Standards? The Role and the Implications of Harmonised Standards for Fundamental Rights Protection in the AI Act

Abstract

The promotion of trustworthy Artificial Intelligence (AI), ensuring a high level of protection of health, safety, and fundamental rights, is the declared objective of the AI Act. However, since the publication of the proposal by the EU Commission and throughout the entire legislative procedure, the indeterminacy of several key provisions has fuelled the scholarly debate. In this context, the presumption of conformity, enshrined in Article 40 of the AI Act, applicable to high-risk and general-purposes AI models complying with harmonised technical standards (HTS), assumes a pivotal function for providers. While the recourse to harmonised standards is not new in EU law, scholars have emphasized numerous controversial aspects of the process of entrusting the determination of harmonised standards to private bodies, i.e. European Standards Organizations (ESOs). The contribution focuses on the role of harmonised standards for fundamental rights protection, scrutinising their relevance within the broader AI Act architecture specifically for high-risk AI systems. Firstly, it examines the relationship between the risk approach to fundamental rights and harmonised standards. Secondly, it explores the adequacy and the implications of the recourse to HTS in the fundamental rights realm, in light of the delegation doctrine. Ultimately, it argues that HTS are structurally ill-suited to ensure effective fundamental rights protection from AI risks.