📄 中文摘要
2月24日，Anthropic 发布了一份详细报告，指责三家中国 AI 实验室（DeepSeek、Moonshot AI 和 MiniMax）对 Claude 进行工业规模的蒸馏攻击。报告中提到，涉及24,000个虚假账户和超过1600万次交换，攻击目标包括推理、工具使用、编码和计算机视觉等能力。尽管地缘政治框架引起了广泛关注，报告的技术内容同样重要，提供了 AI 模型能力提取的杀链分析，对任何通过 API 暴露模型能力的系统构建或防御者都有实际意义。
Anthropic 发布了 AI 模型盗窃的杀链分析，我们来解析一下
📄 English Summary
Anthropic Just Published a Kill Chain for AI Model Theft. Let's Break It Down.
On February 24, Anthropic released a detailed report attributing industrial-scale distillation campaigns against Claude to three Chinese AI labs: DeepSeek, Moonshot AI, and MiniMax. The report highlighted the involvement of 24,000 fraudulent accounts and over 16 million exchanges, targeting capabilities such as reasoning, agentic tool use, coding, and computer vision. While the geopolitical context has garnered significant attention, the technical content of the report is equally important. It presents a kill chain analysis for AI model capability extraction, offering concrete insights for anyone involved in building or defending systems that expose model capabilities through APIs.