The European Union Artificial Intelligence Act (AI Act) represents the first comprehensive, binding regulatory framework for artificial intelligence systems based on a risk-oriented approach. While the regulation provides a detailed set of ex-ante obligations for high-risk AI systems, its implications for real-world deployment—particularly in longitudinal clinical and social care contexts—remain complex and partially underexplored. This paper offers a systematic and article-by-article analysis of the EU AI Act, with a specific focus on risk classification, governance obligations, conformity assessment, and post-deployment monitoring. Particular attention is paid to high-risk use cases in mental health, addiction services, and dual diagnosis settings, where AI systems may influence decision-making processes affecting vulnerable populations over extended time horizons. Without advancing normative or prescriptive positions, the paper identifies structural strengths, ambiguities, and implementation challenges inherent in the AI Act, especially regarding post-inference governance, cumulative risk, and the management of uncertainty in fragile populations. The analysis aims to support policymakers, regulators, and practitioners in understanding the operational consequences of the AI Act in high-risk clinical and social care environments.