The European Union Artificial Intelligence Act (EU AI Act) introduces a comprehensive regulatory framework for high-risk AI systems, particularly in healthcare and clinical decision support. While the regulation emphasizes technical robustness, transparency, and human oversight, it pays limited attention to how sustained interaction with AI systems may reshape human cognition over time. This paper argues that high-risk AI systems can function as external cognitive regulators, progressively influencing users’ decision-making processes, emotional regulation strategies, and sense of agency. Drawing on the extended mind framework and transdiagnostic cognitive-behavioral models, AI reliance is conceptualized as a potential cognitive safety behavior: a strategy that provides short-term relief from uncertainty and decisional burden while posing long-term risks of cognitive narrowing, avoidance, and reduced autonomy. Using high-risk clinical AI systems as a paradigmatic case, with particular reference to addiction care, the paper examines how these cognitive dynamics intersect with the assumptions embedded in the EU AI Act. The analysis highlights regulatory blind spots related to human cognitive adaptation and proposes directions for integrating cognitive and psychological considerations into AI governance and risk assessment.