This paper examines how emerging policy discourses in China, the European Union (EU), and the United States (US) frame artificial intelligence (AI) risks and governance under conditions of scientific and social uncertainty. Drawing from Science and Technology Studies (STS) and framing research, it develops an analytical framework to examine the actors, processes and content of AI risk framing efforts e.g., how AI risks are defined, categorized, assessed and addressed. Findings show distinctive framings from the three jurisdictions, embedded in different dynamics of co-production – how politics, norms and knowledge interact to stabilize the foundations for sense-making. In each case, political considerations and cultural values set the parameters for risk identification and categorization, while experts from industry, academia, and public research institutions carry out risk assessments and response formulations. These complex dynamics generate both convergence and divergence in risk management approaches. Situating these findings in a socio-technical perspective that views AI not as isolated technical artefacts, but as networks/systems of technologies, institutions, regulations, and social practices, this analysis sheds light on the far-reaching implications of these AI risk framings for AI development and associated social responses.