Artificial intelligence is becoming embedded in diagnosis, documentation, communication, risk prediction, and clinical decision support. These uses complicate autonomy because patients may depend not only on a clinician's recommendation but also on models they cannot inspect, organizations that select and monitor those models, data infrastructures that may change, and clinicians whose own knowledge and communication are increasingly mediated by AI. Conventional informed consent remains indispensable, but its episodic emphasis on authorizing a recognizable intervention may incompletely address these continuing dependencies. This article develops informed dependence as a proposed bioethical concept for identifying when autonomy requires attention to the conditions under which patients must rely on AI-mediated clinical arrangements. It distinguishes epistemic, technical, organizational, and relational dependence and argues that materially consequential dependence may generate differentiated duties of disclosure, oversight, continuity, and feasible exit or recourse. The framework does not require patients to understand model mechanics or demand separate consent for every computational process. Instead, it directs attention toward information and safeguards that affect whether reliance is reasonably intelligible, contestable, monitored, and revisable. This account reframes consent as extending beyond a single authorization moment while preserving proportionality and clinical practicability. Its principal limitations are conceptual rather than empirical: the proposed categories and duties have not been validated as a clinical consent model, their relevance will vary across technologies and jurisdictions, and meaningful alternatives may sometimes be unavailable. Informed dependence is therefore offered as a normative lens for analyzing autonomy under sustained AI mediation, not as a universal legal rule or implementation-ready consent standard.
