D1.2 SECURITY THREAT MODELING FOR AI-BASED SYSTEM ARCHITECTURE
The security of machine learning-based systems is not sufficiently addressed at the present time. Methodologies for modelling threats and assessing the security posture of machine learning-based systems are required. This document reviews existing approaches to threat modelling conventional and machine learning-based systems. We identify their limitations and provide improvement directions. Among these solutions, we identify a comprehensive list of vulnerabilities exposed by machine learning-based systems and exemplify how they can be used to infer the extent to which machine learning-based systems are exposed to security threats. We perform threat modelling of both centralized and distributed training and inference paradigms. The result of this analysis enables the identification of fine-grained security requirements for machine learning-based systems.
Read the deliverable here!