US AI leaders turn to Chinese open-weight models ...

The review was performed after leading models from OpenAI and Anthropic refused to help. The article, titled 'US AI leaders turn to Chinese open-weight models,' reports that more American AI leaders are describing Chinese open-weight models as better for AI safety and security than closed-source models.

Ng's account serves as an example of this trend. The article names the two Chinese open-weight models used: Kimi K3 from Moonshot AI and GLM-5.2 from Zhipu AI. The security review concerned OpenWorker, which the article identifies as an open-source AI agent tool. The article notes that the models are Chinese and that the AI leaders are American. It does not specify which OpenAI or Anthropic models refused, nor the exact scope of the review.

The report comes amid an ongoing debate about whether open or closed approaches to AI are safer. The article suggests that the use of Chinese open-weight models for security reviews may affect perceptions of AI safety and security.