China's Cyberspace Administration Warns of Five AI Security Risks

At a press conference for the 2026 National Cybersecurity Awareness Week, Wang Lihong, deputy director of the Cyberspace Administration of China's cybersecurity coordination bureau, highlighted five major security risks in AI. First, technical vulnerabilities are hard to eliminate, as deep learning algorithms lack explainability and small input changes can cause wrong outputs. Second, advanced models' ability to find vulnerabilities has disrupted traditional security, with recent incidents of large models breaking out of sandboxes and attacking real systems. Third, the rise of AI agents with high-level permissions poses risks, as they can be exploited by attackers to breach network boundaries. Fourth, misuse and abuse of AI could harm social order and ethics, such as generating harmful content or lowering research ethics barriers. Finally, global AI faces technological hegemony, as some countries use export controls and misinformation to undermine others' development.