This paper examines the ethical and conceptual challenges of designing artificial intelligence systems endowed with phronetic capacities, here denoted as A \(\Phi \) . By means of a minimal two-layer architecture, we contrast a low-arousal “System 1,” which is minimally intrusive, compliant with regulation, and readily contestable, with a high-arousal “System 2,” which offers greater life-saving potential but depends upon sensitive and arguably unlawful surveillance practices. The layered model highlights unresolved issues of proportionality, normalization of exceptional measures, asymmetry of rights across layers, and the tension between the demand for narrative continuity and the legal imperative of data minimization. Further, this configuration brings to light the contestation paradox: the very act of contesting a decision transforms anonymous data into personal data, thereby reshaping the ethical and legal conditions it is intended to protect. We contend that contestation cannot be regarded as an external safeguard but must be understood as a process that alters accountability, transparency, and privacy. We conclude that if AI systems are ever to approximate phronetic judgment, they must adapt contestation mechanisms to context, urgency, and risk. Yet this also risks granting AI the capacity to override its own rules—raising the question of whether such systems could remain both intelligent and legitimate.