Tag
This literature review identifies and analyzes the problem of silent failures in physical AI systems, where black-box models may execute harmful actions without detection. It proposes a taxonomy of runtime guardrail functions and outlines evaluation requirements for safe autonomous systems.