Zero trust architecture for connected aftermarket devices: An adversarial ML approach to securing OTA updates in automotive edge networks

The proliferation of connected aftermarket devices has introduced critical vulnerabilities in over-the-air update ecosystems for automotive networks. This article presents a novel Zero Trust framework leveraging adversarial machine learning to authenticate device communications without traditional PKI dependencies. The architecture implements behavioral fingerprinting through lightweight LSTM models analyzing CAN bus telemetry and continuous device attestation via edge-based neural networks. Adversarial robustness is achieved through generative adversarial networks synthesizing attack vectors to harden detection models. The middleware-agnostic deployment enables cloud-coordinated model updates while preserving endpoint autonomy. Validated across multiple anonymized aftermarket supply chains, the framework successfully detects supply chain attacks with minimal false positives, adds negligible latency per authentication cycle, and reduces cloud security overhead through edge preprocessing.