Anthropic创始人兼CEO Dario Amodei在周一发表声明，澄清公司从未主张禁止开源权重模型[1]。这一澄清回应了此前Nvidia及Meta、Microsoft、Mistral等AI公司在周五发表的联合信函，该信函敦促政策制定者避免对开源AI模型施加广泛的"过早限制"[1]。
Amodei表示，不危险的开源模型属于公共产品[1]，但他指出开源模型在生物攻击等危险场景中难以施加安全防护[1]。他的核心担忧聚焦于地缘政治风险：专制政府将建立比美国更强大的模型以实现"永久军事优势"[1]。针对这一威胁，Amodei建议的应对措施包括限制中国获取强大芯片和正式打击蒸馏行为[1]。
为应对AI风险，Amodei支持创建全球模型安全测试组织，并主张包括中国在内的所有国家都应参与其中[1]。
Anthropic founder and CEO Dario Amodei issued a statement on Monday to clarify that the company has never advocated for a ban on open-weight models [1]. His remarks came in response to a joint letter released Friday by Nvidia and other AI companies, including Meta, Microsoft, and Mistral, urging policymakers to avoid broad "premature restrictions" on open-source AI models [1].
While defending open-weight models themselves, Amodei emphasized that non-dangerous open-source models serve as public goods [1]. However, he raised concerns about the difficulty of implementing safety safeguards when such models are deployed in dangerous scenarios, such as biological attacks [1]. Amodei's primary concern centers on authoritarian governments developing AI models more powerful than those of the United States to achieve what he termed "permanent military superiority" [1].
To address these risks, Amodei proposed measures including limiting Chinese access to advanced chips and formal efforts to combat model distillation [1]. He also advocated for establishing a global model safety testing organization with participation from all countries, including China [1].