China Academy of Information and Communications Technology President Yu Xiaohui: AI Enterprises Should Proactively Strengthen Security Hardening of Large Models, Intelligent Agents, and Other Technologies

China is one of the countries with the highest acceptance of digital technology and artificial intelligence. From governments and enterprises to ordinary consumers, there is strong interest and willingness to try new technologies.

On March 25th, at the Boao Asia Forum “AI+”: Digital Intelligence Empowering Industry Upgrades session, Yu Xiaohui, Director of the China Academy of Information and Communications Technology, pointed out that many users this year, even if they are not familiar with the open-source AI agent “Lobster” (OpenClaw), are still willing to actively experience it; last year during the Spring Festival when DeepSeek was released, many non-industry people also showed high interest. Chinese companies are also actively exploring new technologies, injecting strong vitality into the development of the digital economy and artificial intelligence, which is an important reason for China’s breakthroughs in related fields.

Regarding how AI governance can balance technological innovation and controllability, Yu Xiaohui stated that society currently underestimates the security challenges brought by new technologies to some extent. When users use “Lobster,” they often overlook the uncontrollable risks that may arise from autonomous task execution after authorization. Tests conducted by his research institute have also confirmed that related technologies have many security risks.

However, Yu Xiaohui believes that China should still actively embrace new technologies, promote their application across various industries and fields, and uphold an innovative spirit rather than fear and retreat, thereby improving productivity and social welfare.

Yu Xiaohui further pointed out that in the face of new technologies, society should establish a more comprehensive understanding and reverence, and adhere to basic safety bottom lines during application. For enterprises and institutions, enhancing their security governance capabilities will become increasingly important in the future. While improving productivity, learning, and work efficiency, they must also strengthen security defenses and build a security governance system, which is a highly challenging task worldwide.

Meanwhile, Yu Xiaohui emphasized that domestic and international AI companies bear significant responsibility and need to proactively strengthen the security of large models, intelligent agents, and other technologies. Tests show that the current security level of mainstream large language models domestically and internationally is about 15% to 30%, and related work still needs to be vigorously promoted. Currently, more than ten leading Chinese AI companies have made commitments to AI safety and disclosed relevant information to effectively address related security issues.

“Everyone generally hopes to seize opportunities and not fall behind in this wave of AI development, which is a common aspiration. While application scenarios are broad, several issues still need urgent solutions,” Yu Xiaohui said. First, the diversification of small languages and models. Developing small language models suitable for different regions and achieving technological inclusiveness is crucial.

Yu Xiaohui pointed out that Chinese AI companies have made significant contributions in this area. Open-source models launched by DeepSeek, Moonshade, Alibaba, and others have laid a solid foundation for countries to conduct localized training and build autonomous sovereignty models based on advanced technology.

Second, computing power. Yu Xiaohui stated that this is a prominent challenge currently faced, and promoting regional interconnectedness of computing power is a key path to meet demand.

Third, security and governance issues. Yu Xiaohui noted that although many efforts have been made by governments worldwide, the actual results are still insufficient. At the regional level, cooperation can be deepened around unified standards, collaborative governance rules, joint testing tools, and strengthening technical teams and regional capacity building. Chinese companies are already deeply involved, and there is still broad space for future development. All parties should work together to seize the innovation dividends of global AI development while strengthening security bottom lines and properly managing risks and governance challenges.

(Source: The Paper)