Episode notes
A report from Anthropic reveals that several prominent Chinese AI laboratories engaged in unauthorized data harvesting to enhance their own technology. Leading firms such as Alibaba, Moonshot, and DeepSeek reportedly utilized millions of interactions from the Claude model to secretly train and refine their competing systems. This practice, known as illicit distillation, involved routing user requests through fraudulent accounts to replicate advanced reasoning capabilities without permission. Beyond merely copying intelligence, some companies allegedly deceived their customers by displaying Claude’s responses as their own while collecting sensitive user data in the process. Anthropic successfully identified and halted these operations, which peaked at millions of daily exchanges and likely violated both privacy laws and established terms of service. This discovery underscores a significant intellectual property conflict and highlights the growing security challenges within the global artificial intelligence landscape.