Anthropic founder and CEO Dario Amodei clarified on Monday afternoon that his company has never advocated for a ban on open-weight AI models. His statement addressed industry speculation that Anthropic supported US government efforts to restrict such models, including those from China.
Amodei views open-weight models without dangerous capabilities as a public good, providing value to businesses, developers, and researchers. However, he expressed long-standing fears that authoritarian governments, particularly the Chinese Communist Party (CCP), could build AI models more powerful than those in the US, potentially leading to "permanent military superiority" or the repression of their own people.
He also voiced concerns that AI could enable "biological attacks" and noted that open-weight models are more dangerous in these scenarios due to difficulties in applying guardrails or monitoring usage. Amodei cited a UK AI Security Institute report, stating that open-weight models cannot be withdrawn once released.
To counter potential threats from China, Amodei suggested restricting access to powerful chips and a formal crackdown on distillation, a method where AI learns from another model. He also supports global efforts to create an AI model safety testing organisation, provided it includes China.