arXiv:2509.00117v1 Announce Type: cross Abstract: The field of embodied AI (EAI) is rapidly advancing. Unlike virtual AI, EAI can exist in, learn from, reason about, and act in the physical world. Given recent innovations in large language and multimodal models, along with increasingly advanced and responsive hardware, EAI systems are rapidly growing in capabilities and operational domains. These advances present significant risks, including physical harm from malicious use, mass surveillance, and economic and societal disruption. However, these risks have been severely overlooked by policymakers. Existing policies, such as international standards for industrial robots or statutes governing autonomous vehicles, are insufficient to address the full range of concerns. While lawmakers are increasingly focused on AI, there is now an urgent need to extend and adapt existing frameworks to account for the unique risks of EAI. To help bridge this gap, this paper makes three contributions: first, we provide a foundational taxonomy of key physical, informational, economic, and social EAI risks. Secondly, we analyze policies in the US, EU, and UK to identify how existing frameworks address these risks and where these policies leave critical gaps. We conclude by offering concrete policy recommendations to address the coming wave of EAI innovation, including mandatory testing and certification for EAI systems, clarified liability frameworks, and forward-looking strategies to manage and prepare for transformative economic and societal impacts.