Sovereignty Theater: Why Sovereign AI Falls Short of Cultural Self-Determination

Abstract

Sovereign AI has emerged as a major policy movement, with nearly 130 government-backed initiatives across more than 50 countries justified in part by the promise of cultural and linguistic self-determination. States seek national AI systems that reflect local languages, values, and knowledge traditions rather than importing the assumptions of foreign technology providers. Yet sovereign AI initiatives overwhelmingly depend on hardware, base architectures, and training methodologies originating from a concentrated set of U.S. vendors, and recent empirical research suggests that the dominant technical approach—fine-tuning a foreign-developed base model on local data—does not reliably produce culturally aligned outputs. We argue that sovereign AI in its current commercial form merely produces sovereignty theater: operational sovereignty (control over compute location and jurisdictional access) without epistemic sovereignty (genuine reflection of local cultural values and knowledge traditions). Bridging the sovereign AI policy literature with empirical research on LLM cultural alignment, and drawing on case analyses of ALLaM (Saudi Arabia) and SEA-LION (Singapore/Southeast Asia), we propose an analytical framework distinguishing infrastructural sovereignty, epistemic sovereignty, and participatory legitimacy. Importantly, our framework formalizes a bottleneck condition in which overall sovereignty is constrained by its weakest dimension. Genuine sovereign AI requires reorienting the project from a commercial product to a participatory endeavor that includes citizens, linguistic minorities, and civil society in the value-laden choices that shape model behavior. We close with a lifecycle decision matrix for policymakers and procurement teams to diagnose sovereignty theater in practice. Although developed in the context of sovereign AI, sovereignty theater offers a portable analytical lens for examining claims to technological sovereignty more broadly.