Our position on open-weights models
In response to recent discussions about open-weights models, especially those from China, Dario Amodei addresses reports that US officials are considering banning US companies' use of Chinese open-weights models. He notes that many tech companies have signed a letter supporting open-weights models and that some have accused Anthropic of wanting to ban them to protect its business. Amodei states clearly that Anthropic has never advocated for a ban on open-weights models, and argues that open-weights models without dangerous capabilities are a public good, providing value to businesses, developers, and researchers at only the cost of compute.
Amodei then outlines his two primary national security concerns. His primary concern is that authoritarian governments—most notably the CCP—could build AI models more powerful than those of the US and use them for permanent military superiority or deep repression of their own people. He notes this concern is widely shared in the US government, citing Vice President Vance's warning and the Intelligence Community's 2026 Annual Threat Assessment. He emphasizes that it is irrelevant whether such models have open weights or are used by US businesses, and that the most dangerous model might be trained in secret and given only to military and security agencies.
His secondary concern is that powerful AI models could be misused for cyberattacks or biological attacks and may have alignment problems. He acknowledges that open-weights models, regardless of origin, present higher risk than closed models because they are difficult to guardrail and cannot be withdrawn once released. However, he argues that banning US businesses from using these models does nothing to address that risk, since bad actors are unlikely to be legitimate US businesses, and that such bans would merely protect US AI companies from competition—which has never been his goal.
The post begins to outline what Amodei does support to address these concerns, but the displayed excerpt cuts off mid-sentence.