EmbedIDS: Open-Source Intrusion Detection for Embedded AI Systems

Summary

EmbedIDS is an Apache 2.0-licensed runtime intrusion-detection SDK for low-power embedded and IoT devices. It uses compact, fixed-logic anomaly-detection models and on-device behavioral analysis to provide real-time protection without cloud connectivity, external services, or dynamic memory allocation. Developers can customize the metrics collected for a device, define detection rules, and train specialized machine-learning models with the accompanying development tools. The SDK is intended for constrained environments where traditional security software may be too resource-intensive or unreliable offline. Its documented use cases include industrial sensors, smart-home devices, medical IoT, automotive systems, and edge-computing nodes. The included EdgeML case study shows a simulated Jetson-class device monitored for GPU hijacking, model tampering, data exfiltration, and telemetry loss. The demonstration reports trust and integrity checks for the device, model, configuration, and agent binary, along with a one-second telemetry heartbeat and an 11-rule monitoring cycle; the page notes that the displayed timings are compressed. The project provides a C11 static library, example code, tutorials, and a GitHub-based quick start, with community contributions and security updates supported under the open-source model.