The growing frequency of malicious attacks on Internet of Things (IoT) devices has rendered conventional approaches with static label-dependent risk assessment models obsolete, especially when coping with unknown and continuously evolving threats. To mitigate these challenges, a novel dynamic trust evaluation framework approach is proposed in this work. The proposed framework utilized unsupervised learning and zero-knowledge proofs to assess device risks in complex environments adaptively, with an accuracy rate of 98.96% for normal clustering and 95.39% for anomalies. K-means clustering algorithm is leveraged to distinguish risk patterns with an additional Decision Tree classification algorithm to analyze the distinguishing characteristics of the behaviors of normal and anomalous devices. The architecture is evaluated in a simulated environment based on real device interaction, with various malicious attacks proportions. In addition, Zero Trust Architecture is integrated into this novel framework to ensure no implicit trust exists between devices, which enforces trust assessment before any collaboration or data exchange.
Lo et al. (2026) studied this question.