Anthropic首席执行官达里奥·阿莫迪(Dario Amodei)发表声明，明确表示"Anthropic从未主张禁止开源权重模型"[1]。这一澄清针对美国官员考虑禁止美国公司使用中国开源权重模型的相关传闻[1]。
阿莫迪阐述了Anthropic对AI国家安全的核心关切[1]。首要威胁是权威政权开发超越美国的AI模型，以获得军事优势或实施深层次压制[1]；其次关切是强大AI模型被滥用于网络或生物攻击[1]。美国副总统万斯在巴黎警告称，"专制政权已经盗取并使用AI来强化其军事、情报和监视能力"[1]。
针对这些威胁，阿莫迪提议采取三项具体措施，而非笼统禁止开源模型[1]。这些措施包括禁止向中国出口先进芯片、打击工业规模的蒸馏操作，以及对所有足够强大的模型进行强制安全测试[1]。他指出，蒸馏技术能使中国利用较少芯片构建更优模型，可能将中国frontier的差距缩小至美国frontier的数个月内[1]。
Anthropic CEO Dario Amodei has issued a statement to dispel misconceptions about the company's position on open-weights AI models.[1] Amodei explicitly stated that "Anthropic has never advocated for a ban on open-weights models," addressing reports that U.S. officials were considering prohibiting American companies from using Chinese open-source model weights.[1]
Rather than supporting outright restrictions, Amodei articulated two primary national security concerns regarding artificial intelligence.[1] The first concern centers on authoritarian regimes developing AI systems more powerful than those of the United States, which could grant them military advantages or enable large-scale repression.[1] The second concern involves the potential misuse of advanced AI models for cyberattacks or biological attacks.[1] Vice President Vance has previously warned that "authoritarian regimes have stolen and used AI to strengthen their military, intelligence, and surveillance capabilities."[1]
To address these threats, Amodei proposed three targeted measures instead of a blanket ban on open-weights models.[1] These include prohibiting the export of advanced semiconductors to China, disrupting large-scale model distillation operations, and implementing mandatory safety testing for all sufficiently powerful models.[1] Amodei noted that distillation techniques could enable China to develop superior models than chipset limitations would typically allow, potentially narrowing the technological gap between China's frontier capabilities and America's by several months.[1]