Andrew Ng argues that Chinese open-weight models offer superior safety and transparency compared to restricted U.S. closed-source systems.
AI pioneer Andrew Ng stated that open-weight models are safer and more transparent than closed-source models, challenging the common belief that open-source systems present a significant threat to society. During the Agentic AI Summit, Andrew Ng explained that while leading U.S. models from OpenAI and Anthropic often refuse to assist with potentially harmful tasks due to strict safeguards, these restrictions can sometimes block legitimate requests, such as those needed to improve cybersecurity. To demonstrate this, Andrew Ng and a colleague utilized Chinese models, specifically Moonshot AI’s Kimi K3 and Zhipu AI’s GLM-5.2, to conduct a security review of their new tool, OpenWorker. These Chinese models were chosen because they did not refuse the tasks assigned to them, unlike the U.S. models which were often too restrictive. The debate highlights a growing divide between frontier capabilities and safety practices. While closed models offer controlled environments, open-weight models allow users to run the technology on their own hardware, where they can modify any safeguards. Experts suggest that the objective should be making high-quality, safe capabilities accessible to everyone while systematically removing the most hazardous risks.