Anthropic指控阿里巴巴利用25000个虚假账户对Claude实施史上最大规模模型蒸馏攻击
英文摘要
Anthropic has sent a confidential letter to the US Senate alleging that Alibaba orchestrated the largest recorded AI distillation attack. The operation reportedly used 25,000 fake accounts to extract and clone the capabilities of Anthropic's Claude model. The accusation marks the first publicly known instance of such a large-scale model extraction attempt. The brief report did not disclose further details on methods or evidence.
中文摘要
Anthropic向美国参议院提交了一封机密信函，指控阿里巴巴策划了有记录以来最大规模的AI模型蒸馏攻击。据称该行动使用了25,000个虚假账户来提取并克隆Anthropic的Claude模型的能力。这是首次公开的如此大规模模型窃取指控，但简短报道未披露具体手段或证据。
关键要点
Anthropic accused Alibaba of deploying 25,000 fake accounts to distill and clone Claude's capabilities.
Anthropic指控阿里巴巴部署25000个虚假账户来蒸馏并克隆Claude模型的能力。
The accusation was made in a confidential letter sent to the US Senate.
该指控是在一封提交给美国参议院的机密信件中提出的。
This is described as the largest-ever recorded AI distillation attack.
这被描述为有记录以来最大规模的AI模型蒸馏攻击。