Anthropic opposes open model ban; NVIDIA assembles 75-signature alliance in 4 days

On 27 July, Anthropic CEO Dario Amodei published a blog post responding to the open letter 'Open Weights and American AI Leadership,' explicitly stating Anthropic never advocated banning open-weight models. He argued such models without dangerous capabilities are a public good. The post broke four days of silence that had worked against Anthropic. The context included Moonshot AI releasing Kimi K3 with 2.8 trillion parameters on 16 July, and escalating US government rhetoric against Chinese open models. Amodei proposed three alternative measures: no chip sales to China, clamp down on industrial distillation, and mandatory safety testing for all sufficiently capable models, both open and closed, with exemptions for startups and academic use. The first measure targets NVIDIA, which has been selling chips to China under a special arrangement. Meanwhile, NVIDIA CEO Jensen Huang had posted an open letter on 24 July supporting open models, gathering 75 signatures by 28 July, including Microsoft, Meta, IBM, Dell, Palantir, Mistral, Hugging Face, and later OpenAI, Google, and SpaceX. Anthropic and Amazon did not sign. On 27 July, NVIDIA launched the Open Secure AI Alliance and announced a $5 billion investment in Ilya Sutskever's Safe Superintelligence (SSI). The article notes that the real battle is over who gets to decide what constitutes a 'frontier model,' with Anthropic favoring an industry body (like Demis Hassabis's proposed Frontier AI Standards Body) and NVIDIA opposing such restrictions to protect its GPU market.