The Universality Myth: Epistemic Exclusion, Structural Power, and AI Governance in the Global South

<title>Abstract</title> <p>The OECD AI Principles and the EU AI Act are supposed to apply to all; the UNESCO Recommendation is also touted as being universal. This paper suggests that the claim is ideological and not descriptive as most of the populations it purports to represent do not have contestation literacy, which is the ability to see an AI system as an object of governance and to evaluate a compliance claim against it, as opposed to usage literacy which the deployer has a vested interest in developing. This is the systematic lack of conceptual resources that allows to understand, challenge and influence the AI systems that dominate communities in the Global South. The paper combines document and policy analysis with Vallor's AI mirror theory, Fricker's and Dotson's epistemic injustice frameworks, Jasanoff's sociotechnical imaginaries, and Lukes's and Gaventa's structural power theory to compare India (elite capture and regulatory self-exemption) with sub-Saharan Africa (foreign-built systems adopted without institutional capacity), showing the pattern is structural, not exceptional. Five contributions follow. First, epistemic exclusion is a functional mechanism of structural power, self-reinforcing because closing it would expose AI deployers' compliance claims to scrutiny they avoid. Second, Vallor's mirror framework is extended: within her own scope for populations who perceive but misread the mirror, and through Fricker and Dotson for populations who cannot perceive it at all. Third, an internal mechanism of AI power supplements decolonial AI critique's predominantly external focus. Fourth, Jasanoff's sociotechnical imaginaries explain how science-fiction fear and techno-utopian narratives alike render structural AI harms imaginatively unavailable, structuring rather than prohibiting the question. Fifth, the unawareness/misinformation distinction extends to the facilitator enabling deployment: documented evidence from India's infrastructure-clearance process (Section 6.4) shows bureaucratic actors structurally excluded, by their role's categorical architecture, from the governance system they enable. Generalising this mechanism beyond the case examined here remains a falsifiable extension for future work.</p>