During the 'May Day' holiday, robots and robot dogs, as typical examples of embodied AI devices, were a common sight in tourist attractions and shopping malls across various regions, underscoring a flourishing industrial landscape. Yet, this surge has also brought to light safety concerns. At the 8th C3 Security Conference in Chengdu, Sichuan, a striking demonstration showcased a commercial robot dog being remotely hijacked via non-contact network techniques. This highlighted the potential dangers of embodied AI devices when they malfunction, including the covert activation of cameras, unauthorized collection and transmission of data, and even attacks on dummy models. An iResearch report predicts that the embodied AI market will witness trillion-dollar demand in the coming decade. However, the current industry is grappling with the dilemma of having 'robust hardware but feeble software,' lacking the necessary capabilities to address real-world challenges effectively. In light of these safety risks, various stakeholders are actively seeking to establish a comprehensive security protection system. This includes issuing standard frameworks and promoting the development of national standards. Security firms are also making tangible efforts, such as employing a 'model-governed-model' strategy and introducing products like AI XDR 2026. Furthermore, fostering industrial collaboration and nurturing talent are vital pillars in overcoming these safety hurdles.