Embodied AI systems, including humanoid and quadruped robots, are being deployed in various settings, leveraging large language models and robotics advancements to perform complex tasks autonomously. However, security measures have not kept pace, leaving these systems vulnerable to exploitation. Researchers have successfully hijacked commercially available robots via Bluetooth, exfiltrating sensitive data to servers in China and creating physical botnets by infecting neighboring robots wirelessly. This raises significant concerns, particularly given the increasing presence of these robots in critical infrastructure and military deployments1. The development of large language models, particularly those originating from China, is reshaping the capability and risk landscape of embodied AI. As a result, security implications are emerging as a critical consideration, underscoring the need for robust security protocols to mitigate potential threats. This matters to practitioners because the lack of adequate security measures in embodied AI systems can have severe consequences, including compromised data and disrupted operations.