The AI Era Begins, Urging a New Paradigm for Digital Security
-
On August 9, the 'ISC 2023 11th Internet Security Conference' themed 'Security as a Service: Opening a New Paradigm for Digital Security in the AI Era' commenced in Beijing. Attendees highlighted that data security issues in the AI era are particularly pressing, requiring joint efforts from technology and regulation to address the security challenges of large AI models. They emphasized the need for forward-looking integration of AI and cybersecurity technologies to enhance the resilience and security of industrial and supply chains.
AI Model Security Becomes a Hot Topic
Currently, the security of large AI models has emerged as a critical issue affecting the industry's development. Luo Fengying, Deputy Director of the Cybersecurity Coordination Bureau of the Cyberspace Administration of China, noted that while generative AI significantly boosts economic and social progress, challenges such as the accuracy and authenticity of generated information remain unresolved. Academician Wu Jiangxing of the Chinese Academy of Engineering pointed out that large AI models face inherent security issues, including 'three intractabilities'—unexplainability, undecidability, and non-deducibility—as well as common vulnerabilities like backdoors.
Wu Jiangxing explained that AI systems, composed of hardware, data systems, algorithms, and applications, inherently possess security flaws. The 'black box' nature of neural networks leads to unexplainable AI behavior, while over-reliance on training data results in undecidable outcomes. Additionally, the irreversible nature of neural network processes makes deductions impossible. These issues, combined with hardware vulnerabilities, create unprecedented challenges for governments and enterprises.
Data Security Concerns in the AI Era
Academician Wu Hequan, Honorary Chairman of ISC, stressed that data security is a major concern in the AI era. Large models rely on vast datasets, often sourced from open or internet data, raising quality and content risks. Using proprietary corporate data also introduces leakage risks. Luo Fengying added that transforming data into economic value depends not just on volume but on accessible, shareable data for businesses and individuals.
Wu Hequan acknowledged that while companies and national centers conduct AI security assessments, many risks remain undetected due to limited understanding of AI's security implications.
Collaborative Innovation for Security Solutions
Industry experts called for joint efforts between enterprises and regulators to tackle AI model security. Cybersecurity firms and AI developers are key players in addressing these challenges. Zhou Hongyi, founder of 360 Group, argued that traditional security approaches are insufficient; cross-disciplinary solutions are needed, leveraging expertise in both AI and cybersecurity.
Wu Jiangxing proposed using Dynamic Heterogeneous Redundancy (DHR) to enhance AI system security. The Purple Mountain Laboratory will host a global challenge in Q4 2023, focusing on AI system security, inviting experts to collaborate on solutions.
Regulatory Measures and Market Growth
Luo Fengying announced the 'Interim Measures for the Management of Generative AI Services,' effective August 15, aimed at balancing innovation and regulation. The rules target public-facing AI services, excluding research and development activities.
Shang Bing, Chairman of the Internet Society of China, advocated for collaborative innovation, urging leading firms to pioneer security technologies while startups focus on niche innovations. Compliance and governance were also emphasized.
Expanding Cybersecurity Market
China's cybersecurity market is growing, with 2022 software revenues reaching $3.92 billion, up 12.5%. The sector is robust, with 21 listed firms and increasing participation from tech giants like Alibaba and Tencent in AI security. Zhou Hongyi revealed 360 Group's development of a security-focused AI model, aiming for 99% accuracy in threat detection.
Sui Jing, Director of the Ministry of Industry and Information Technology's Cybersecurity Bureau, highlighted innovation as key to security advancement, urging support for 'specialized and sophisticated' firms and fostering a dynamic industry ecosystem.