Anthropic CEO Dario Amodei has rejected calls to ban open-weights models, stating the company has never advocated for such restrictions. In a statement, Amodei emphasized that open-weights models without dangerous capabilities are a public good, offering value to businesses, developers, and researchers. He argued that banning these models does not address the most serious national security concerns, including the risk of authoritarian governments developing AI with military or surveillance capabilities.
Amodei outlined two primary national security concerns. First, he warned that authoritarian regimes, particularly the Chinese Communist Party, could build AI models more powerful than those developed in the U.S., potentially achieving permanent military superiority or enabling deep repression. Second, he highlighted the risk of powerful AI models being misused for cyberattacks or biological threats. He noted that open-weights models pose a higher risk because they are harder to monitor or control once released. However, he argued that banning their use by U.S. companies does little to mitigate these risks, as malicious actors are unlikely to be legitimate businesses.
Amodei reiterated that Anthropic supports three key measures to address these concerns: restricting chip exports to China, cracking down on industrial-scale distillation operations, and requiring mandatory safety testing for all sufficiently capable models, regardless of their openness. He emphasized that safety testing is the most effective way to address risks like cyber and biological threats, and that such testing should be global, including cooperation with China. Source: anthropic
