AI Analysis: Threat-Oriented Digital Twinning for Security Evaluation of Autonomous Platforms

Unlocking Secure Autonomy: Threat-Oriented Digital Twinning

This paper presents a novel methodology for cybersecurity evaluation of learning-enabled autonomous platforms using a threat-oriented digital twin. The approach focuses on architectural trust assumptions, verifiable assurance tests, and open-source implementation to study spoofing, replay, and adversarial ML stress. It demonstrates how higher-layer assurance mechanisms can bound operational effects even when transport authentication is compromised, highlighting the value of explicit separation between perception, decision, and control.