Beyond the Rubber Stamp: When Is Human Oversight Meaningful?

JOTTERAND, Alexandre. Beyond the Rubber Stamp: When Is Human Oversight Meaningful? Five Operational Conditions for AI Systems and Automated Decisions under the FADP, GDPR and AI Act. In: MÉTILLE/TAMÒ-LARRIEUX (éd.), L'IA et la protection des données personnelles. Berne: Stämpfli, CEDIDAC, 2027, p. 35–63.

Abstract (AI generated)

Swiss and European law converge on the demand that AI and automated systems must remain subject to human oversight. Article 21 of the Swiss Federal Act on Data Protection, article 22 of the EU General Data Protection Regulation, article 14 of the EU Artificial Intelligence Act and article 8 of the Council of Europe Framework Convention on Artificial Intelligence all express, in different registers, the same expectation. In practice, the expression human in the loop has become a compliance shorthand that often masks the absence of any genuine intervention. The human is present, but control is not. Conversely, inserting a human into the loop may degrade outputs rather than improve them. This article identifies the operational baseline common to all four legal frameworks and examines when each form of human involvement — human-in-the-loop (HITL), human-on-the-loop (HOTL), human-in-command (HIC) and human-out-of-the-loop (HOOTL) — is appropriate and under what conditions any of them can deliver meaningful control. Drawing on the Meaningful Human Control (MHC) and Meaningful Human Oversight (MHO) frameworks, the article identifies five operational requirements, namely authority, competence, information, enabling environment, and fitting intentions, that form a common baseline across the four instruments and should, in my view, anchor Switzerland's forthcoming AI legislation.