Generative AI, Health Misinformation, and Medical Deepfakes: Implications for Diagnostic Safety and Clinical Practice—A Narrative Review

Generative artificial intelligence (AI) can rapidly produce persuasive health information in the form of text, synthetic images, cloned voices and deepfake videos. It can also be used to create fictitious professional identities. This narrative review describes how synthetic claims, fabricated evidence and simulated identities are used to disguise misleading health information and give it the appearance of authenticity especially with regard to diagnostic safety and clinical practice. Literature identified through Web of Science and Scopus was synthesized across text image audio video and multimodal formats; current evidence indicates that generative AI can produce highly credible but inaccurate health content. Source authority, message framing and professional identity cues influence trust and verification behavior, while disease-specific evaluations show that high average accuracy may coexist with clinically important omissions, missed referrals, hallucinated findings and poor calibration. In medical imaging, errors may arise both from incorrect interpretation and from manipulation of the diagnostic evidence itself. Detection, evidence-based verification, provenance, literacy interventions, and governance provide complementary safeguards, although real-world and longitudinal evidence remains limited. Clinical evaluation should therefore consider task-specific performance, source authenticity, human oversight, and the consequences of individual errors alongside overall accuracy.