A Study on Extending the TARA Methodology for Physical AI: A Case Study of ADS

Automated Driving Systems (ADS) integrate complex Physical AI, rendering traditional software-centric threat modeling inadequate against emerging AI and physical-layer vulnerabilities. We propose an extended Threat Analysis and Risk Assessment (TARA) methodology, compliant with ISO/SAE 21434, that unifies STRIDE, MITRE ATLAS, and OWASP Top 10 for LLMs. Driven by a newly constructed Attack Library, our framework systematically captures multi-layered cyber-physical threats. Applied to an ADS case study on the NVIDIA Jetson platform, the method yields 38 security checklists. These outputs cover all 32 UN R155 Annex 5 Part A threat categories, identify 29% more AI-specific risks than a STRIDE-only baseline, and achieve ML 3 for data-source integration under D2TARA. Our key contributions are: an integrated Physical AI TARA Methodology, a comprehensive Attack Library synthesized from CVEs, CWEs, and Papers, and 38 actionable security checklists validated against UN R155 Annex 5.