A Framework for Evaluating Global AI Governance Initiatives

ABSTRACT Efforts to develop global governance initiatives for artificial intelligence (AI) have increased significantly in recent years. However, these initiatives have generally had a limited impact due to their vagueness, lack of authority and repetition. Several factors contribute to the difficulties in establishing effective global AI governance mechanisms, including geopolitical tensions, institutional gridlock, and the general‐purpose and sociotechnical characteristics of AI. Developing politically legitimate governance mechanisms that can operate within these constraints and effectively compel behaviour change among government and industry actors is essential for building a more mature global AI governance ecosystem. In this article, we contribute to this aim by introducing a framework for evaluating the political legitimacy of global governance initiatives. It is designed to clarify why many global AI governance initiatives lack authority and to identify opportunities for more impactful international cooperation. We operationalise the framework by assessing global AI governance initiatives which address two international security problems: establishing regulation for lethal autonomous weapon systems and implementing safety testing for general‐purpose AI.