Proposing a five-layer taxonomy for physical AI security

Researchers present a common framework that unifies the risks in physical artificial intelligence systems, ranging from hardware security to semantic integrity. This classification aims to guide future security research by identifying gaps in the fragmented literature.