Carnegie Endowment Senior Fellow Discusses China's AI Open-Weight Model Policy Amidst Security Concerns
Graham Webster, a lecturer and research scholar at Stanford University’s Center for International Security and Cooperation and editor-in-chief of the DigiChina Project, analyzes China's approach to open-weight AI models. Despite Chinese President Xi Jinping's recent call for 'open source and openness, cooperation and sharing' in AI development, there's an underlying tension with the need for 'security and controllability.' This duality, enshrined in Chinese legal texts, suggests that while China embraces open models, it also prioritizes security. Carnegie Endowment for International Peace senior fellow Matt Sheehan notes that 'control' extends beyond AI systems going rogue to include broader disaster prevention. Chinese companies like DeepSeek, Z.ai, and Moonshot continue to release open-weight models, offering lower-priced, customizable competition to closed models from companies like OpenAI and Anthropic. The Chinese government has not yet barred these releases, indicating that officials do not view them... 
