How Knowledge Graphs Are Reinventing AI Security: Insights from ISC.AI 2025

The 13th ISC.AI 2025 Knowledge Graphs Reshaping Intelligent Security Summit was held at the Beijing National Convention Center, organized by the China Computer Federation’s Computing Security Committee, the Chinese Association for Artificial Intelligence’s AI and Security Committee, and the Beijing Chamber of Commerce’s AI Industry Committee, gathering top scholars and industry leaders.

Yan Ming, former director of the CSCC and former head of the Ministry of Public Security’s research institutes, emphasized that the rapid evolution of AI and large models is mutually reinforced by knowledge graphs, improving model accuracy, interpretability, reasoning, and trustworthiness, thereby advancing AI systems and their security.

Yin Yuhui, senior vice‑president and CEO of 360 Group, highlighted knowledge graphs as a core AI technology that offers powerful data integration, deep semantic understanding, and efficient reasoning, enabling full‑dimensional monitoring of data flows and precise identification of security risks.

Wang Xin, associate dean of Tianjin University’s School of AI, presented new KG technologies for the era of large models, noting that multimodal KGs combined with multimodal large models drive industry‑wide standard updates and digital transformation.

Zhang Huaping, professor at Beijing Institute of Technology, discussed the ChatBIT project, showing how KG‑driven domain knowledge graphs empower large models to achieve deeper intelligence and even autonomous adversarial platforms.

Zhang Ruchong, professor at Beihang University, described the “LLM+KG Intelligent Dual‑Engine” concept, explaining how KGs assist large‑model training and fine‑tuning, breaking complex problems into reasoning chains that enhance inference quality.

Zhang Yanfeng, professor at Northeastern University, stressed that integrating KGs with AI deepens graph neural networks, improving risk control and finding broad applications in Chinese enterprises.

Liu Zheli, dean of Nankai University, argued that foundational large models must align with socialist core values, using KGs to build a value‑aligned corpus that guides domestic model development.

Zhao Yanyan, professor at Harbin Institute of Technology, warned that multimodal large models require safety alignment, embedding human safety principles into language models and establishing evaluation frameworks that manage KG‑LLM security relationships.

The summit concluded that constructing a global knowledge network enables precise threat tracing and proactive warning, reinforcing data security for the digital economy and showcasing the profound impact of knowledge‑driven security.

During the event, three whitepapers were released: the Data Security Joint Operation Technical Whitepaper, the Industrial Internet Enterprise Data Security Joint Control System Product Whitepaper, and the Global Advanced Persistent Threat Analysis Whitepaper, all aimed at advancing KG‑based cross‑domain collaboration and high‑quality industrial development.