Abstract

The implementation in the EU of the CoE Framework Convention on AI, human rights, democracy and the rule of law, will happen through the pre-existing AI Act, complemented by other relevant Union acquis. This contribution argues that the AI Act is not the right instrument to implement the Framework Convention. Besides the fact that the AI Act predates the Framework Convention, the AI Act is a product safety regulation: because product safety should be limited to hazards that can be addressed through technical and procedural measures, it is not a good fit for human rights protection. Additionally, the strong emphasis on meta-regulation and the role of technical standards, the absence of rights for individuals, and the scarce, public law remedies offered by the AI Act, do not make it a good candidate to implement a human rights convention.

TABLE OF CONTENTS

1. Introduction
   1.2 The implementation of the Framework Convention in the EU digital aquis

2. The AI Act: nature, structure, and basic mechanisms
   2.1 The aims and nature of the AI Act
   2.2 A prohibition… ex post?
   2.3 Much ado about risk management and conformity
   2.3.1 The Conformity Assessment
   2.3.2 Quality Management System (QMS)
   2.3.3 Fundamental Rights Impact Assessment
   2.4 Remedies: don’t put off to tomorrow’s law what can be done in today’s law

3. Implementing the Framework Convention via the AI Act: reasons why the circle isn’t squared
   3.1 Product safety is not for everything
   3.2 Remedies and art. 14 of the Framework Convention
   3.3 Market Surveillance Authorities, Fundamental Rights Authorities, and Fragmentation

4. Final reflections and conclusion