Edge AI hinges on integrating lightweight models with specialized hardware—such as smartphones, wearable devices, Internet of Things terminals, and humanoid robots—allowing closed-loop operations (perception, reasoning, and execution) to occur entirely on the device. Photo: IC PHOTO

---

On March 16, scholars convened in Beijing for a seminar on the industrial development and security governance of “artificial intelligence (AI) on the edge,” or “edge AI,” hosted by the Institute of Law at the Chinese Academy of Social Sciences (CASS). At the event, Zhou Hui, head of the incubation project team of the AI Security Governance Research Lab at CASS and deputy director of the cyber and information law research office at CASS’s Institute of Law, announced the release of a research report on the development and governance of edge AI.

Unique advantages

Edge AI refers to the technology that directly deploys the computing and decision-making processes of AI algorithms locally on edge devices, enabling real-time data processing, low-latency response, and privacy protection by reducing dependence on cloud infrastructure. This technology hinges on integrating lightweight models with specialized hardware—such as smartphones, wearable devices, Internet of Things terminals, and humanoid robots—allowing closed-loop operations (perception, reasoning, and execution) to occur entirely on the device.

Compared with cloud-based AI, edge AI boasts distinct advantages. Zhou noted that because edge AI processes data locally, it avoids the risks of leakage and abuse associated with uploading data to the cloud, thus safeguarding user privacy to the greatest extent. Additionally, its rapid real-time response eliminates network delays by processing data on the device itself, meeting the strict real-time demands of scenarios like autonomous driving. Furthermore, with minimal dependence on network connectivity, edge AI can continue to operate AI algorithms and maintain basic functions even in environments with no or weak internet access, thereby improving system reliability.

With the rapid development of AI technology and industry, edge AI devices have become increasingly integrated into everyday life, as a relatively clear hierarchical collaboration system within AI applications has already taken shape. Zhang Yan, deputy dean of the Law School at Renmin University of China, explained that AI applications today have formed a complete chain comprising four levels: general capabilities, domain-specific applications, intelligent execution, and terminal implementation—specifically, general large models, vertical large models, AI Agents, and edge AI terminals. Each level corresponds to a different industrial entity, each bearing distinct obligations and responsibilities.

Zhang emphasized that the healthy development of the edge AI industry requires legal frameworks tailored to the characteristics of edge AI, as well as a clear delineation of rights and responsibilities among these four levels, in order to effectively protect the legitimate interests of end users.

China’s responses to challenges

In recent years, China has introduced a series of policies at both national and local levels to support the development of the AI industry. For example, the Ministry of Education’s “AI Empowerment Action” explicitly calls for the deep integration of intelligent technologies with education, research, and society, providing policy support for the application of edge AI in areas such as education and healthcare.

The Implementation Opinions on Promoting the Innovative Development of Future Industries, jointly issued by seven ministries including the Ministry of Industry and Information Technology, further clarifies support for the research, development, and application of intelligent terminal products, offering policy guidance for the deployment of edge AI in sectors like consumer electronics and industrial manufacturing.

Moreover, cities such as Shenzhen, Guangdong, and Wuhan have introduced targeted policies to spur technological breakthroughs and the industrialization of edge AI.

Nevertheless, the development of edge AI faces multiple challenges, including dependence on cloud infrastructure for computing power and energy consumption, issues related to model reliability and compatibility, and heightened risks stemming from data sensitivity.

Zhou told CSST that while edge AI’s local data processing reduces the risk of data leaks—a key advantage—it does not eliminate security concerns. On the contrary, since edge devices process users’ sensitive data directly, such as facial recognition information, voice recordings, and health data, the risks escalate if devices are attacked, misused, or lack clear authorization protocols. For example, smart wearable devices like smartwatches and health monitors typically collect physiological data, including heart rate, sleep patterns, and location. If this data is maliciously exploited, it could result in severe infringements of user privacy.

Though edge AI’s growth raises risks in privacy, security governance, and data access boundaries, existing laws like the Cybersecurity Law, Data Security Law, and Personal Information Protection Law provide a foundation to address these issues. However, these regulations require further refinement to tackle new scenarios and applications.

Hu Naying, a senior business director from the China Academy of Information and Communications Technology (CAICT), stressed that addressing the challenges posed by edge AI requires an approach rooted in industrial practices, analyzing issues by entity, scenario, and tier to foster effective governance while promoting industrial development. At present, CAICT is actively exploring risk governance for edge AI. Technologically, efforts are focused on building a comprehensive AI agent security framework that defines security risks, attributes, and requirements to lay a strong groundwork for agents’ safety architecture.

Edited by CHEN MIRONG