Governing Open Large Language Models in public administrative decision-making: a framework for accountability, justification, and institutional control

Abstract

The integration of Open Large Language Models (OLLMs) into administrative decision-making raises critical challenges related to accountability, justification, and institutional control. This paper examines OLLMs not merely as technical tools but as governance objects embedded in legally sensitive decision processes. It identifies the administrative justification problem as the gap between algorithmically plausible outputs and institutionally defensible decisions. To address this, the study proposes a governance framework structured around three dimensions: accountability, administrative justification, and public institutional control. The framework introduces a four-stage cycle (authorization, justification design, oversight, and audit) intended to help institutions require, structure, and assess—rather than guarantee—that AI-supported decisions remain traceable, reviewable, and compliant with administrative standards. As a conceptual framework, it has not been empirically validated and is illustrated through a worked example rather than tested. The paper contributes by reframing OLLM adoption as a governance problem and offering a structured approach to preserving legitimacy and procedural fairness in public and institutional contexts.