Towards a Methodological Approach for Understanding Human Oversight Requirements under the AI Act

ABSTRACT Human-centric AI has become a central principle in AI regulation, embodied in Article 14 of the EU AI Act, which requires that meaningful human oversight be designed into high-risk AI systems from the outset of their development. This paper examines Article 14 from a methodological perspective and proposes three core steps for analyzing its legal requirements: (1) identifying and categorizing obligations across text-based, soft, embedded, and datadriven law; (2) mapping the relational aspects of responsibility across the AI supply chain; and (3) xamining how these obligations are implemented across the system lifecycle. By systematically addressing the layered nature of legal requirements, relational responsibilities, and the timing of rule implementation across the lifecycle of AI, the paper offers a basic framework for understanding how human oversight can be meaningfully and effectively embedded within the design, development, and operation of AI systems.

KEYWORDS Human oversight - AI Act - Proactive law - Data-driven law - Distributed responsibility

TABLE OF CONTENTS 1. Introduction. – 2. Human oversight under the AI Act as design-based regulation. – 3. Towards a methodological approach to HoBD. – 3.1. Identify legal requirements in written law and beyond. – 3.1.1. Law 1.0 – Black letter law. – 3.1.2. Law 2.0 – Soft law. – 3.1.3. Law 3.0 – Embedded law. – 3.1.4. Law 4.0 – Data-driven law. – 3.2. Identify responsible actors as well as the relations between them. – 3.2.1. Technology providers. - 3.2.2. Deployers. – 3.2.3. Endusers. – 3.2.4. Regulators. – 3.3. Distributed esponsibility. – 3.3.1. Design stage. – 3.3.2. Development stage. – 3.3.3. Evaluation stage. – 3.3.4. Operation stage. – 3.3.5. Retirement stage. – 4. Conclusion.