Wang Lihong, a senior official at the central Cyberspace Administration of China (CAC), told state broadcaster CCTV recently that the five main AI safety/security risks today are: 1. Inherent technical vulnerabilities i.e. deep learning algorithms that lack explainability; biased training data that lead to discriminatory outputs 2. Frontier models disrupting traditional security fields i.e. vulnerability discovery and exploitation at scale, sandbox escapes during safety evaluations 3. Evolving risks from autonomous agents i.e. OpenClaw and other "high privilege" agents that can become “Trojan horses” for cyberattacks 4. Misuse and societal impacts i.e. harmful outputs that raises risks of bioweapon development; labor displacement 5. Tech hegemony by "certain countries" i.e. export controls; "misleading AI evals" and wielding of influence over public discourse to malign Chinese AI