U.S. AI Leader Shifts to Chinese Open-Source Models for Security Audits

Foreign Media: Leaders in U.S. AI are shifting toward Chinese open-source weight models, challenging the security claims of closed-source models.

AI pioneer Andrew Ng stated at the California Summit that open-source weight models are safer than closed-source ones. When reviewing his open-source AI tool, OpenWorker, models from OpenAI and Anthropic refused to assist. Ultimately, the security review was completed using Moonshot AI's Kimi K3 and Zhipu AI's GLM-5.2. Previously, Hugging Face also used GLM-5.2 to defend against network attacks launched by OpenAI models.

Ng and others believe that the excessive safety mechanisms in recent U.S. models such as GPT-5.6 and Fable 5, while intended to prevent harmful activities, also hinder legitimate cybersecurity requests. In contrast, Chinese open-source models demonstrate superior performance in cybersecurity and transparency.

Original article: toutiao.com/article/1872671453670400/

Disclaimer: The views expressed in this article are solely those of the author.