Physical AI Governance Framework Proposed Across Five Lifecycle Stages

A recent publication on arXiv (2607.22877) explores the governance issues unique to Physical AI—systems that engage with and navigate the physical environment. Unlike conventional AI, these systems must adhere to real-time safety protocols, continuously interact with changing surroundings, and operate alongside humans. The authors compile existing governance concepts into a cohesive framework specifically for Physical AI. They introduce a five-phase lifecycle—research, design, data, model development, and deployment—and illustrate how governance can be effectively applied at each phase with practical implementation strategies. This paper highlights deficiencies in current AI governance models, which fail to address the distinct risks associated with Physical AI.