3家中国人工智能公司使用24 000个假账户, 通过未经授权的进入, 窃取其聊天室的能力。 U.S. firm Anthropic alleges three Chinese AI companies used 24,000 fake accounts to steal its chatbot’s capabilities via unauthorized access.

美国AI公司人类学组织指控三家中国公司DepSeek、Moonshot AI和Minimax使用约24 000个假账户与其Claude Chatbot进行1 600多万次未经授权的互动,据称是为了通过模型蒸馏获取先进能力。 U.S. AI company Anthropic has accused three Chinese firms—DeepSeek, Moonshot AI, and MiniMax—of using around 24,000 fake accounts to conduct over 16 million unauthorized interactions with its Claude chatbot, allegedly to extract advanced capabilities through model distillation. 该活动违反了服务条款和区域准入规则,目的是通过复制Claude的推理、编码和安全防患于未然的对策,培训自己的AI系统。 The activity, which violated terms of service and regional access rules, aimed to train their own AI systems by copying Claude’s reasoning, coding, and safety-resistant responses. 人类通过IP数据和基础结构模式查明了这些公司,注意到它们迅速适应新的模型更新。 Anthropic identified the firms through IP data and infrastructure patterns, noting rapid adaptation to new model updates. 继OpenAI提出类似指控后, 事件突显了在AI竞争、知识产权及国家安全方面日益紧张的关系。 The incident highlights growing tensions over AI competition, intellectual property, and national security, following similar allegations by OpenAI. 中国公司没有回应。 The Chinese companies have not responded.