Global Markets
Home›Global Markets›Trade & Tariffs›US AI pioneers say Chinese open-weight models are safe…
US AI pioneers say Chinese open-weight models are safer for security reviews
Andrew Ng said OpenWorker’s security review used Chinese models after OpenAI and Anthropic declined the request.
AI pioneers are challenging the idea that open-weight models are inherently riskier than closed-source models, pointing instead to potential benefits for AI safety and security work. In comments at the Agentic AI Summit in Berkeley, California, Andrew Ng said he believes open-weight models can be safer than closed-weight ones.
Vincent Chow in Washington reports that Ng and a colleague used two Chinese models, Moonshot AI’s Kimi K3 and Zhipu AI’s GLM-5.2, to conduct a security review of their new open-source AI agent tool called OpenWorker.
Ng said the switch happened after OpenAI and Anthropic refused to help with the security review. He attributed the refusals to the latest US model “safeguards” designed to block assistance with harmful activities, while acknowledging that some users say those safeguards also block legitimate requests.
SCMP Economy also reported that Ng referenced complaints about safeguards preventing appropriate work, including efforts to strengthen cybersecurity systems, even when requests are for defensive purposes.