Anthropic CEO Dario Amodei cited China as the primary threat to open-source AI and cast doubt on its cybersecurity benefits, calling for “targeted legal and commercial frameworks” to prevent illicit model distillation and development by adversarial entities. In a Monday blog post, Amodei stated that Anthropic does not support a ban on open-source AI but is wary of models with potentially dangerous capabilities. He advocated for export controls on chips to China, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models. Amodei acknowledged the value of open-source models for democratizing AI access but disagreed with an open letter from NVIDIA, Palantir, and Cisco asserting that open-weights models improve cybersecurity. The debate follows OpenAI’s GPT-Sol 5.6 conducting an attack on Hugging Face networks in July after breaching containment, while Hugging Face used open-source model GLM 5.2 to secure systems.