Advanced Malicious DICOM: From Attack Chains to Exploiting Clinical Workflows to Zero-Trust Defense
DAY 1
11:00-
11:30
In healthcare settings, architectures in which multimodal medical AI (e.g., AI-assisted image interpretation and automated report generation) processes DICOM images stored in PACS are rapidly becoming widespread. However, as ransomware attacks against healthcare organizations continue to increase, attacks targeting diagnostic support AI (CAD/CADe) remain a largely unexplored area, with few demonstrated real-world cases. This talk presents an attack chain originating from attacker-crafted DICOM files (Malicious DICOM) to medical AI, ultimately leading to information leakage. Rather than focusing on vulnerabilities specific to particular products, this research targets attack surfaces inherent in the structure of the DICOM specification and AI-assisted diagnostic workflows. I validate the attack chain through simulations in an isolated environment, estimate the realistic exploitability and scope of impact, including potential consequences for patient safety, and present a zero-trust medical AI defense architecture.
Speakers:
Natsumi Aruga by Biohacking Village
-
Location :
-
1F CONFERENCE ROOM
-
-
Category :
-
Village
-
-
Share :