Deepfake X-rays expose a new medical imaging security gap

The newest warning in radiology is not that AI might miss disease, but that AI can help fabricate convincing evidence of it. Research highlighted by the Radiological Society of North America found that deepfake chest X-rays were difficult for radiologists to distinguish from authentic images, and in some cases also deceived AI systems trained to detect pathology.

That matters because imaging has long benefited from an assumption of evidentiary integrity. Clinicians may debate interpretation, but they generally trust that the pixels originated from the patient and scanner described in the record. Once that assumption weakens, radiology inherits a cyber-physical risk: image tampering could affect insurance claims, disability determinations, emergency triage, clinical trial eligibility, or even targeted fraud against specific patients.

The broader implication is that medical AI evaluation can no longer focus narrowly on sensitivity, specificity, and workflow speed. Vendors and health systems will need to think in terms familiar to cybersecurity teams: watermarking, chain-of-custody, scanner-level signing, anomaly detection, and audit trails that link images to acquisition context. PACS and reporting systems may need to evolve from storage tools into authenticity-verification layers.

For policymakers and hospital leaders, this is a reminder that generative AI risk in medicine is not confined to chatbots hallucinating text. Clinical media itself can now be manipulated in ways that are plausible enough to influence care. Radiology may become one of the first specialties where AI safety, cybersecurity, and medico-legal governance fully converge.