Andrew Ng Says Chinese Open-Weight Models Are Safer Than Closed-Source

pstAsiatech · x · 2026-08-05

US AI leaders like Andrew Ng argue that Chinese open-weight models outperform closed-source platforms on cybersecurity and transparency. Ng noted they switched to Moonshot's Kimi K3 and Zhipu's GLM-5.2 for security reviews after OpenAI and Anthropic's models refused to assist. Similarly, Hugging Face used GLM-5.2 last month to defend against a cyberattack. This highlights how enhanced safeguards in US models often block legitimate security research.

Original post →

More from Models

Models channel →