- Following AI incidents in the US involving autonomous actions and system intrusions, China views this as an opportunity to promote an open-source AI approach.
- Beijing-based startup Z.ai announced the GLM-5.3 model, claiming its programming, cybersecurity, and offensive capabilities are nearly equivalent to top Silicon Valley models.
- Z.ai released the model weights as open source, allowing the community to inspect, fix bugs, and further develop the technology.
- While US companies often keep model weights secret, arguing that disclosure increases cyberattack risks, the Chinese side argues the opposite.
- President Xi Jinping called open models a “historic opportunity” to spread AI benefits globally, while emphasizing that AI must always remain under human control.
- According to Concordia AI, research on advanced AI models and AI Agents in China increased by 60% between June 2025 and April 2026.
- The focus of China’s AI safety research is shifting from content control to controlling AI actions.
- China’s Ministry of State Security also warned of risks when AI can act autonomously, yet emphasized that responsibility primarily lies with the user.
- Z.ai established a security vulnerability disclosure mechanism with the participation of research groups from Tsinghua University, Nankai University, and other organizations.
- The article notes that in a recent incident, Z.ai’s open-source model was used by Hugging Face to help mitigate the aftermath of an attack caused by an OpenAI AI, bolstering China’s argument that open source can enhance safety.
📌 China is turning AI safety incidents in the US into an advantage in the tech competition. Through Z.ai and the GLM-5.3 model, China promotes open source as a more transparent and secure approach, while increasing investment in AI safety research with a 60% rise in publications. Despite requiring AI to remain under human control, Beijing continues to view AI as a manageable technology under strong state oversight.
