AI MODEL GOVERNANCE

This Position Paper, developed within the AI Model Governance Department of the Italian National Institute for Artificial Intelligence (ENIA) and coordinated by Director Stefano Romano, explores the critical issue of model risk and the role of governance frameworks in the era of Artificial Intelligence. The paper defines the conceptual foundations of AI Model Governance, distinguishing between traditional models and AI-based models, and examines the unique risks associated with their design, implementation, and use. Particular emphasis is placed on the multidimensional nature of model risk, including data bias, methodological fragility, operational weaknesses, and ethical and reputational concerns. The study maps the international regulatory and standardization landscape, with a specific focus on the EU AI Act (Reg. 1689/2024), ISO/IEC 42001:2023, and the NIST AI Risk Management Framework, highlighting their implications for organizations adopting AI systems. The paper also outlines governance structures (first, second, and third lines of defense), lifecycle management practices, and tools such as model registers, risk tiering, and explainability frameworks. By adopting a multidisciplinary perspective—integrating engineering, law, economics, sociology, and public administration—the document argues that AI Model Governance should not be reduced to a compliance exercise. Instead, it must be understood as a proactive infrastructure of collective trust, supporting transparency, accountability, and sustainable innovation. Ultimately, the paper positions model governance as a public function: a critical element in shaping responsible AI ecosystems where human oversight, ethical safeguards, and institutional resilience converge to protect rights, enhance trust, and foster innovation.