Anthropic has accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of using model distillation to extract over 16 million responses from its Claude chatbot, allegedly violating its terms of service and undermining U.S. export controls. The labs reportedly used around 24,000 fraudulent accounts for industrial-scale extraction to train competing AI systems. Anthropic’s claims have received criticism and mockery online, with many pointing out that Anthropic and other major AI companies themselves train models on publicly available data without consent, fueling a broader debate over intellectual property, copyright, and fair use in AI development. Distillation is a technique where a smaller model learns from a larger one's outputs; in this context, it is seen as a method for extracting and replicating proprietary capabilities. Anthropic warns that such practices are becoming more frequent and pose risks beyond any single company, calling for coordinated industry and regulatory action. Meanwhile, Reddit has sued Anthropic for allegedly scraping its content to train Claude; this case is part of broader legal disputes over AI training data practices. Anthropic is implementing countermeasures but emphasizes that collective action is needed to address distillation attacks.