Adversarial Machine Learning Threats To Medical Device AI Controllers
Author(s): Venkata Sai Abhinav Piratla
Publication #: 2604012
Date of Publication: 07.03.2026
Country: United States
Pages: 1-15
Published In: Volume 12 Issue 2 March-2026
DOI: https://doi.org/10.62970/IJIRCT.v12.i2.2604012
Abstract
The integration of artificial intelligence into life-critical medical device controllers—including closed-loop insulin delivery systems and cardiac monitoring devices—introduces adversarial machine learning (AML) attack surfaces that conventional cybersecurity frameworks do not adequately address. Adversarial attacks targeting these systems carry direct patient safety implications, yet no comprehensive, medical-device-specific AML threat taxonomy exists in the current literature. This paper addresses that gap by presenting a structured taxonomy of 34 adversarial attacks spanning training-phase, inference-phase, privacy, and model integrity threat categories, with explicit analysis of applicability to resource-constrained medical hardware. Building on prior work in artificial pancreas security [1] and autoencoder-based anomaly detection [2], a catalog of 25 defense mechanisms is evaluated against the computational and memory constraints typical of implantable, wearable, and portable device classes. Identified threats are mapped to the MITRE ATLAS adversarial machine learning framework and aligned with FDA cybersecurity and AI lifecycle guidance, providing a regulatory reference for device manufacturers and premarket submission planning. Tiered defense recommendations are derived from device resource profiles, enabling practitioners to select contextually appropriate countermeasures. The taxonomy and defense landscape together constitute an actionable resource for engineering adversarially resilient AI-enabled medical devices.
Keywords: Adversarial machine learning, medical device security, threat taxonomy, artificial pancreas, MITRE ATLAS, evasion attacks, data poisoning.
Download/View Count: 11
Share this Article