'Not against open-weight AI models': Anthropic CEO clarifies stance
What's the story
Anthropic founder and CEO Dario Amodei has clarified his company's position on open-weight models. In a blog post, he emphasized that "Anthropic has never advocated for a ban on open-weights models." This statement comes amid industry speculation about his company's support for US government efforts to restrict such models, particularly those from China.
Industry response
NVIDIA's Jensen Huang shared an open letter on X
The clarification from Amodei comes after NVIDIA founder Jensen Huang shared an open letter on X.
The letter, signed by several leading AI companies including Meta, Hugging Face, Microsoft, Mistral, and NVIDIA, urged policymakers to refrain from imposing broad "premature restrictions" on open-weight AI models.
The industry debate has largely focused on concerns that Chinese AI labs are advancing their capabilities by allegedly stealing intellectual property from American counterparts through techniques like distillation.
Clarification
Distinction between open-weights models and China threat
In the blog post, Amodei differentiated between open-weight models and the potential threat posed by China.
He said, "Open-weights models that don't have dangerous capabilities are a public good: they don't cost anything besides the compute needed to run them, and they provide value to businesses, developers, and researchers."
Despite his fears about AI in general, he doesn't consider businesses using these models, even those from China, a threat.
Concerns
Amodei's concerns about authoritarian governments
Amodei's fears are mainly centered around "authoritarian governments" developing more powerful models than those in the US for "permanent military superiority."
He also worries about these regimes using AI to suppress their own people.
The Chinese Communist Party (CCP) is not the only authoritarian government he fears, but it's the "most capable," he said.
AI risks
AI could facilitate biological attacks, Amodei fears
Amodei also expressed fears that AI could facilitate "biological attacks," not just cybersecurity ones.
He believes open-weight models are more dangerous in these scenarios as it's hard to enforce guardrails or monitor their usage.
He cited a UK AI Security Institute report that once released, open-weights can't be withdrawn, which goes against the argument of open-source advocates who believe access to powerful models helps protect defenders.
Safety measures
Call for global model safety testing organization
Amodei also expressed support for the establishment of a global model safety testing organization, especially if all countries including China agreed to it.
He said, "I have been heartened both that the Trump administration has moved in this direction in recent months, and by recent industry proposals that would apply such testing to the most capable models regardless of their country of origin or whether they are open or closed."