From Abstract Threats to Institutional Realities: A Comparative Semantic Network Analysis of AI Securitisation in the US, EU, and China
Artificial intelligence governance exhibits a striking paradox: while major jurisdictions converge rhetorically around concepts such as safety, risk, and accountability, their regulatory frameworks remain fundamentally divergent and mutually unintelligible. This paper argues that this fragmentation cannot be explained solely by geopolitical rivalry, institutional complexity, or instrument selection. Instead, it stems from how AI is constituted as an object of governance through distinct institutional logics. Integrating securitisation theory with the concept of the dispositif, we demonstrate that jurisdictions govern ontologically different objects under the same vocabulary. Using semantic network analysis of official policy texts from the European Union, the United States, and China (2023-2025), we trace how concepts like safety are embedded within divergent semantic architectures. Our findings reveal that the EU juridifies AI as a certifiable product through legal-bureaucratic logic; the US operationalises AI as an optimisable system through market-liberal logic; and China governs AI as socio-technical infrastructure through holistic state logic. We introduce the concept of structural incommensurability to describe this condition of ontological divergence masked by terminological convergence. This reframing challenges ethics-by-principles approaches to global AI governance, suggesting that coordination failures arise not from disagreement over values but from the absence of a shared reference object.