Qianxin Han Yonggang: The development and application of large-scale models has brought new security risks, and AI security is still in its infancy.

At the 2025 China Internet Conference, Han Yonggang, Vice President of Qihoo 360, delivered a speech. He stated that the security of AI itself is very different from traditional security, and currently the focus of protection is mainly on AI research and development testing environments, AI-related data, and applications in the security field, which is still in its early stages. In terms of content security, the challenges in the cognitive field, as well as the difficulty in the future in terms of intelligent body permission control, application and data protection, are still high, and this is also an increment of the future network security industry. AI is an increment for both the supply and demand sides of network security. It is expected to transform small-scale high-level capabilities into large-scale offerings, drive the industry from labor-intensive to intelligence-intensive, and improve the efficiency of the security industry. He mentioned that the development and application of large models bring new security risks. The black box properties of large models, connections to various businesses and types of people, application to multi-dimensional data, and the lack of effective security assessment, protection, monitoring measures, and methods during the rapid deployment process all pose significant security risks in large model applications; AI security not only involves traditional security issues but also new issues such as content security.