<p>The dominant framework for analyzing artificial intelligence centeres on risk—the probability that systems will malfunction, be misused, or produce unintended consequences. We argue that this framing systematically obscures what matters most: the susceptibility of people, institutions, and infrastructures to harm. Using proceedings from the Yale Digital Ethics Center’s Digital Vulnerabilities in the Age of AI Summit (DIVAS), we advance a complementary reframing, from AI risk to AI-mediated vulnerability. We identify five mechanisms through which AI transforms existing vulnerabilities—speed, scale, scope, asymmetry, and opacity—and map their manifestation across four analytical levels: individual, relational, institutional, and infrastructural. The analysis reveals cross-cutting dynamics—a widening gap between technological dependency and democratic control, the limits of technosolutionism, and a persistent cat-and-mouse trap that frustrates technical countermeasures—that explain why interventions targeting a single level or domain consistently fall short. We defend three normative implications for governance: context-specific regulation matched to vulnerability dynamics, inversion of the burden of proof from affected parties to AI deployers, and investment in the social conditions—community institutions, knowledge ecosystems, trust relationships—that technosolutionist approaches neglect. AI governance cannot be approached as a purely technical exercise; it is ultimately a question of political power, burden allocation, and democratic accountability.</p>