Secure Tiny Machine Learning on Edge Devices: A Lightweight Dual Attestation Mechanism for Machine Learning

Vlad-Eusebiu Baciu, An Braeken, Laurent Segers, Bruno da Silva

Emerging edge devices are transforming the Internet of Things (IoT) by enabling more responsive and efficient interactions between physical objects and digital networks. These devices support diverse applications, from health-monitoring wearables to environmental sensors, by moving data processing closer to the source. Traditional IoT systems rely heavily on centralized servers, but advances in edge computing and Tiny Machine Learning (TinyML) now allow for on-device processing, enhancing battery efficiency and reducing latency. While this shift improves privacy, the distributed nature of edge devices introduces new security challenges, particularly regarding TinyML models, which are designed for low-power environments and may be vulnerable to tampering or unauthorized access. Since other IoT entities depend on the data generated by these models, ensuring trust in the devices is essential. To address this, we propose a lightweight dual attestation mechanism utilizing Entity Attestation Tokens (EATs) to validate the device and ML model integrity. This approach enhances security by enabling verified device-to-device communication, supports seamless integration with secure cloud services, and allows for flexible, authorized ML model updates, meeting modern IoT systems’ scalability and compliance needs.