Allocating Responsibility for Open-Source Agentic AI: The Construction of the Regulated Entity in EU Technology Law
Year: 2026
Type: dissertation
Abstract: Every obligation in the EU Artificial Intelligence Act is addressed to an identifiable person. Open-source agentic frameworks do not supply one. This thesis asks how EU law decides whom to place obligations on when no producer of an AI system can be identified, and why the AI Act did not do so here. Taking OpenClaw, its add-on registry and its US foundation as a most-likely case, it reads the Act actor by actor and weighs nine allocation devices across eight instruments. No actor in the middle of the chain holds a determinate status, and in a sanctions-backed regime that uncertainty engages the interest the legality principle protects. Where a status does attach, the duties that follow are not aimed at the conduct that produced the harm. The systemic-risk regime is keyed to model capability, so its duties land on the model's provider rather than on the deployed population. One architectural choice produces all three. The Cyber Resilience Act, facing the same difficulty, built an addressee: the open-source software steward. Why they diverged is put as a question and tested; a deliberate policy choice about open source is best supported and still incomplete. Other instruments narrow the gap without closing it, and what none requires of this chain is the screening of what a registry publishes, vulnerability handling and coordinated disclosure. The contribution is doctrinal and comparative: the technique exists and was not applied here, and whether it should be is left open.
Author: VASQUEZ-PACHECO, JUAN JOSE
Language: Italian
Subfield: Law
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7220268296