Clinical cyberbioethics for AI-mediated clinical decision-making: a critical narrative review and conceptual proposal for an encounter-level deliberative model

Background Artificial intelligence (AI) increasingly influences diagnosis, prognosis, triage, treatment, and resource allocation. Existing ethical and governance frameworks identify important values and lifecycle duties, but institutions still lack a shared encounter-level procedure specifying who must act, what must be checked and documented, and when an AI-supported decision should be overridden, paused, or escalated. Objective To map recurrent ethical and governance requirements in a purposively assembled source corpus and use the mapping transparently to construct an encounter-level deliberative model without presenting the model as an automatic finding of the literature. Methods We conducted a critical narrative review with conceptual mapping of a heterogeneous source set identified through iterative scoping between January and March 2025. The review is not systematic, exhaustive, or representative. The master set contained 54 records; 11 duplicates and one non-evidence record were removed, leaving 42 retained sources. Thirty-five informed the primary conceptual mapping and seven served foundational, methodological, post-search, or companion-publication roles. Three members of the review team, including ADP, participated in source selection; the preserved records do not support a claim of independent triple screening or a retrospective agreement statistic. SWiM was removed. Selected ENTREQ transparency items and SANRA were used only as reporting and self-audit aids. Literature-derived findings, integrative inferences, and author-developed normative proposals were explicitly separated. Results Five recurrent domains were mapped: accountable clinical authority and meaningful human oversight; clinically meaningful explanation, uncertainty communication, and contestability; equity, bias, and distributive impact; lifecycle safety, security, and accountability; and proportional data governance, privacy, and patient agency. These domains informed, but did not prove, the author-developed seven-step Deliberative Clinical Cyberbioethics Model (D-CCM). Each step specifies accountable actors, mandatory inputs and records, four possible outputs (continue, continue with safeguards, pause/stop, or escalate), and escalation triggers. The model also includes five candidate patient-facing protections, explicit trade-off rules, four forms of drift or divergence, and scheduled and trigger-based revision. Comparison with WHO, UNESCO, NIST AI RMF, and FUTURE-AI positions the D-CCM as a complementary point-of-care translation layer. Conclusions Clinical Cyberbioethics is presented as a proposed integrative governance lens, not as an established new discipline. The D-CCM is a practice-oriented conceptual proposal informed by the mapping and requires prospective co-design, feasibility testing, reliability assessment, and outcome evaluation before it can be treated as a validated operational framework.