Recently, DeepSeek has been continuously attacked by hackers, which indirectly reflects the security dilemma faced by domestic large-scale models. It cannot be denied that the rise of open source big models in China, represented by DeepSeek, not only reduces the cost of training and deploying big models, accelerates the popularization of artificial intelligence technology, but also breaks the dual constraints of technology blockade and commercial barriers, promoting the artificial intelligence industry towards a more open new era. However, like a double-edged sword, DeepSeek, as an open source model, also needs to face the common security challenges of all open source software: due to the open source code, the security issues of open source software are more complex than those of closed source software, and attackers can freely view and modify the source code, thereby triggering potential security vulnerabilities. This is not groundless worry, DeepSeek's experience confirms this pattern: on the one hand, external network attacks have a long duration, great destructive power, multiple means, and fast changes; On the other hand, vulnerabilities discovered during deployment and application expose various security risks. In view of this, since the beginning of 2025, various security companies have been on high alert, exploring new paths and paradigms to build a strong network security defense line and safeguard domestic large-scale models. For example, the big model security solution jointly released by AsiaInfo Security and AsiaInfo Technology can be summarized as "escorting artificial intelligence with artificial intelligence". It not only integrates the rich practical experience of AsiaInfo Technology's big model products in key industry fields, but also relies on AsiaInfo Security's research achievements and technical strength in the field of big model security to create a security solution that meets both practical needs and the development requirements of big models. It provides protection through four major systems: cloud security protection system, big model security gateway, terminal data protection system, and AI application risk protection system, establishing a four layer deep protection covering the basic layer, model layer, tool layer, and AI application layer for big models. At the same time, the program also provides two professional services, namely artificial intelligence risk assessment and artificial intelligence security services. Regular risk assessment and evaluation are conducted to safeguard the security of large models and applications. In the future, as more and more application scenarios are implemented, domestic large models will still face endless new risks. The security of large-scale models should not only be a matter of "repairing the lost sheep", but also of being proactive and restructuring the security system. In this regard, multiple interviewed experts expressed that attention should be paid to the "systematic shortcomings" of large models in security architecture, vulnerability response, data compliance, and other aspects. Starting from multiple levels and comprehensively using various means, a comprehensive and multi-level security protection network should be created, especially the construction of the "native security" capability of large models. More importantly, the application potential of large models in the field of security is enormous. With their excellent reasoning ability, customized training methods, and open-source features, they can effectively improve the efficiency and accuracy of various security tasks. Therefore, exploring innovative applications of big models in the field of security and using big models to govern big model security issues has become an important direction. The industry has formed a consensus that the big model is a cross industry ecosystem covering multiple industries. The comprehensive governance of big model security risks is not just a technical issue, but requires the participation of multiple forces such as the government, enterprises, and research institutions to work together, consolidate the foundation of domestic big model development, and open up ideas to safeguard the artificial intelligence industry. (New Society)