MedSTRIDE-AI: An Extended STRIDE-Based Threat Modeling Framework for Multimodal Agentic AI Systems in Medical Internet of Things (MIoT) Environments

Multimodal agentic AI in Medical Internet of Things (MIoT) systems poses compound cybersecurity threats such as intrusion and zero day anomaly, that are not adequately captured by traditional threat models based on STRIDE. This paper proposes MedSTRIDE-AI framework, a structured STRIDE extension developed using Design Science Research (DSR) and grounded in a comprehensive systematic review. MedSTRIDE-AI explores three contemporary intrusion detection system (IDS) threat anomalies in MIoT ecosystem in this research work: Multimodal Adversarial Fusion Attacks (T1.2), Unsafe Autonomous Clinical Action Chains (T4.2), and Cross-Device Cascading Memory Poisoning (T6.2), together with an ISO 14971:2019-inspired patient harm prioritization layer and a six-step threat identification workflow. MedSTRIDE-AI adds a risk-aware IDS response which emphasizes containment based on the magnitude of harm for the patient, thereby reducing the harm-weighted residual risk by 62% more than a harm-agnostic IDS response. The evaluation results show that MedSTRIDE-AI provides not only broader structural threat coverage than STRIDE, MEDSTRIDE, and ASTRIDE, but also measurable security properties and mitigation trade-offs for multimodal agentic healthcare systems.