This paper examines the evolution of Human-Machine Interaction (HMI) through the lenses of legal theory, political epistemology, and governance studies, arguing that the shift from algorithmic profiling to conversational agents and reasoning systems has created unprecedented disruptions in cognitive autonomy, social trust, and normative authority. By critically engaging with key unresolved issues affecting both the safe development of AI systems and AI policy-making processes, the analysis draws on Post-Normal Science (PNS) to call for replacing reactive risk management with anticipatory governance and extending co-regulation beyond industry-led standards. The paper further introduces the concept of digital vulnerability as a transversal legal category designed to capture the epistemic and democratic challenges embedded in AI-mediated interactions.