World Internet Conference AI Committee Focuses on Agent Security, 360 Participates in Governance as a Core Member

On November 2, the 2026 Plenary Meeting and Third Chairmen's Meeting of the Artificial Intelligence Professional Committee of the World Internet Conference were held in Wuzhen, Zhejiang. Pan Jianfeng, Chief Scientist and Senior Vice President of 360 Group, attended the meeting as Deputy Director of the AI Professional Committee. The 2027 work plan (draft) reviewed at the meeting proposed that the committee's security and governance advancement plan will further focus on "controllable and collaborative governance of agent security," driving AI security governance to extend further toward agent behavior control, governance capacity building, and global governance mechanisms. As a member unit of the committee, 360's sustained research and practice in the field of agent security are highly consistent with this direction.

Image: Scene of the AI Professional Committee meeting

As AI moves from content generation to autonomous execution, agents have begun to invoke tools, access data, and autonomously complete tasks, and security risks have extended from the content level to permission management and behavior control. In response to this change, the committee plans to carry out thematic discussions and joint research in 2027 on directions including agent permission management, autonomous decision-making and human control, multi-agent collaboration security, and full-lifecycle security management.

Around these topics, 360 has continued to participate in the committee's related work. In 2026, the committee's security and governance advancement plan conducted on-site research visits to member units including Huawei, 360 Group, Alibaba Cloud Intelligence Group, and Baidu; in July, Pan Jianfeng delivered a keynote speech at the seminar on "Collaborative Governance of Frontier AI Security Risks" held by the committee. These exchanges and research are also built on 360's long-term technical practice.

For agent security, 360 adopts the protection approach of "using models to govern models and using agents to counter agents," and has launched the 360 Agent Security Protection and Management System. The system covers key links such as agent asset discovery, behavior control, threat detection, incident response, and loss recovery. It can identify and manage agents, skill plugins, and model links, and implement runtime control over behaviors such as file access, network connections, and command execution, helping enterprises bring agent operation processes into security management.

It is reported that agent security is only part of 360's AI security system. In the broader field of AI security, 360 has also laid out key directions such as large model security and AI application governance. Among them, its self-developed AI security attack and defense system "Tulongfeng" passed the China Academy of Information and Communications Technology's "Internet Agent Governance - Vulnerability Discovery Agent Capability Assessment," becoming the first product to pass this assessment; at the same time, 360 has participated in the development of multiple AI security national standards and industry standards, forming complete capability support from products to standards.

As agents move from "being able to answer questions" to "being able to execute tasks," AI security governance is moving from discussion of principles to specific management of permissions, behaviors, and execution processes, and the focus of governance is shifting from "managing content" to "managing behavior." 360 will combine its accumulated experience in artificial intelligence and digital security to continue exploring technical paths for controllable agent security and provide practical references for industry governance.

Leiphone Leiphone