Several leading figures in American artificial intelligence are increasingly endorsing Chinese open-weight AI models as safer and more effective for security purposes than US closed-source alternatives. This challenges the prevailing stance of US developers like Anthropic, who argue that open-source models pose societal risks.

At the Agentic AI Summit in Berkeley, California, on Saturday, AI pioneer Andrew Ng, former head of Google Brain and former chief scientist at Baidu, stated, “From what I’m seeing, I think open-weight models seem safer to me than closed-weight models.” Ng and a colleague recently used two Chinese AI 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, OpenWorker, released last month. This step was taken after leading US models from OpenAI and Anthropic declined to assist with the task.

This instance exemplifies a growing trend of Chinese AI models supporting US AI security efforts. For example, last month, the New York-based open-source code repository Hugging Face also employed GLM-5.2 to defend against a cyberattack originating from OpenAI’s models.

The reliance on Chinese models stems from enhanced safeguards in the latest US AI systems, such as OpenAI’s GPT-5.6 and Anthropic’s Fable 5, which are designed to refuse assistance with harmful activities. However, these safeguards have faced criticism for also blocking legitimate requests, including those related to cybersecurity reinforcement.

This development signals a shift in the AI community’s approach to balancing openness, safety, and security in artificial intelligence technologies.

Sources