Recently, the 7th Cloud Security Alliance Greater China Conference (CSA GCR Conference) was held in Shenzhen, China, with the theme of "Cloud Integration for the Future, Security Within". Baidu Vice President Chen Yang attended the conference and delivered a keynote speech on "Security Big Model" and "Big Model Security". Chen Yang stated that while the big model accelerates industrial upgrading and economic growth, it also brings new opportunities and challenges to the cybersecurity industry. Based on the big model, the entire security business process can be reconstructed through intelligent human-machine collaboration, and the big model can evolve into a "productivity tool".
Artificial intelligence is an important driving force for a new round of technological revolution and industrial transformation, and the rise of large models has further expanded the depth and breadth of artificial intelligence applications. Since Baidu released the knowledge enhancement big language model ERNIE Bot in March, the basic model of ERNIE Bot has iterated to the 4.0 version of Wenxin big model, which has significantly improved the four basic abilities of understanding, generation, logic and memory. At the meeting, Chen Yang shared the ideas and practices of building AI native applications in various fields such as security operations, content auditing, APP privacy compliance detection, and business risk control, using the Wenxin Big Model in the security field, helping to significantly improve security efficiency and capabilities.
Chen Yang stated that improving the efficiency of industry AI native security application research and development is also crucial. At present, based on Baidu's leading AI big model platform, Baidu Security has created a Baidu AI security base, a big model based on security knowledge and skill enhancement, and a full scenario intelligent agent. For various security business scenarios, human-machine collaborative security workflows are implemented in a low code manner, and through the AI native security service layer, various security applications can undergo AI native restructuring.
On the other hand, the security, trustworthiness, and controllability of the big model itself have become crucial. Network and system security, model algorithm security, data security and privacy protection, and content security are intertwined throughout the entire lifecycle of the big model. System security and endogenous security will coexist for a long time.
As a leading enterprise in artificial intelligence, Baidu attaches great importance to the capacity building and ecological construction of large model security risk prevention. Based on the practical experience of Baidu's big model security, Baidu Security has summarized a set of response strategies from the perspective of service providers from the perspective of the entire lifecycle of big models. The plan covers key stages such as big model training/fine-tuning/inference, big model deployment, and big model business operation. It provides security products and services throughout the entire process, including data security and privacy protection, model protection, content security, and business operation risk control. At the same time, it combines the idea of using attack to promote defense to establish AIGC content security evaluation capabilities, and implements routine security assessments of big models to help enterprises build stable, healthy, trustworthy, and reliable big model services.
For example, in the operation phase of the big model, Baidu Security has built a series of security services including the red line knowledge base/red line big model, prompt review and rewriting, multimodal content review, content security evaluation, and so on. In response to negative and high-risk red line issues such as illegal activities and negative values, we have focused on building a "red line knowledge base and red line big model" to provide safe response content for answering, and can provide positive guidance to users. In response to Prompt injection attacks, the large model can output secure content by rewriting Prompt while keeping its initial instructions and data unchanged.
It is reported that as a global member of the Cloud Security Alliance CSA, Baidu has long been committed to safeguarding the development of the industry with safer AI, actively exploring and practicing cutting-edge technologies in intelligent cloud computing, and continuously deepening the integration and innovation of "artificial intelligence+cloud security". At the meeting, as the Director of the Artificial Intelligence Security Committee of the Beijing Artificial Intelligence Industry Alliance, Chen Yang represented the committee and officially signed a strategic agreement with the CSA Cloud Security Alliance. In the future, both sides will leverage their respective strengths to carry out joint research and develop cutting-edge technologies and applications in the field of artificial intelligence.
Faced with the new situation, opportunities, and challenges faced by the current "big model+security", Baidu Security will continuously expand the technological innovation and application of artificial intelligence technology in the field of network security. On the other hand, it will maintain cooperation with various sectors, jointly build a big model security defense line, and explore the infinite possibilities of empowering thousands of industries with safer big models.