Anthropic CEO Amodei wrote an article to clarify his position: he does not advocate banning open source, but must block computing power, distillation and pre-release testing. In response to rumors that relevant departments of the US government are considering banning domestic companies from using open source AI models from specific countries, Anthropic co-founder and CEO Dario Amodei publicly stated

Anthropic CEO Amodei wrote an article to clarify his position: he does not advocate banning open source, but must block computing power, distillation and pre-release testing. In response to rumors that relevant US government departments are considering banning domestic companies from using open source AI models from specific countries, Anthropic co-founder and CEO Dario Amodei publicly issued a statement, clearly drawing a line: Anthropic has never proposed or advocated a ban on open source models. Recently, discussions about banning open source models have caused violent reactions in the industry. Many technology companies have signed an open letter to support the open source model. Some people even questioned Anthropic's attempt to use administrative bans to suppress opponents and protect their own business interests. Amodei clarified that as long as the open source model does not have dangerous capabilities, it is a very valuable public product - it is low in cost and can bring significant impetus to enterprises, developers and researchers. Protectionist bans cannot address real national security concerns, but may become a tool for industry monopoly, which is by no means the position or goal of Anthropic. In the article, Amodei broke down his real concerns about the safety of AI technology and national security into two layers. The first level is the risk of authoritarian regimes using AI systems that are more powerful than the United States to gain a permanent military advantage or deepen internal surveillance. He pointed out that in the face of such risks, whether the model has open source weights and whether American companies use these models are not key factors. The most dangerous ones are often secret models developed in secret and only provided to the military or intelligence agencies. The second level is the risk of powerful AI models being maliciously abused to launch large-scale cyber attacks or develop biological weapons. Although open source models lack continuous security protection and monitoring methods and cannot be taken back once released, which indeed poses higher security governance challenges, simply prohibiting compliant local companies from using open source models will not stop malicious non-compliance participants at all, and therefore will not be of any real help in reducing such threats. In order to truly catch the above risks, Anthropic put forward three implementable policy suggestions and regulatory propositions. First, strictly limit the flow of key computing power to potential threat parties - continue to implement export controls on high-performance chips and process equipment, severely crack down on all types of illegal smuggling and OEMs that bypass the ban, and cut off the computing power foundation for high-risk model training at the source. Secondly, crack down on large-scale cross-border "model distillation" (Distill) behavior. Some organizations are using low-cost methods such as distillation to mass-absorb the capabilities of cutting-edge closed-source models, significantly shortening the research and development cycle. In response to this practice of using abusive terms to circumvent computing power blockades, clearer legal and business punishment mechanisms should be established. Finally, implement comprehensive and mandatory pre-r