[Anthropic Claims DeepSeek and Other AI Labs Conduct Large-Scale 'Distillation' of Its Model Capabilities] According to an announcement by Anthropic, it has discovered that DeepSeek, Moonshot AI, and MiniMax have conducted large-scale industrial 'distillation attacks' to extract the capabilities of the Claude model. These labs are accused of using over 24,000 fake accounts to interact with Claude more than 16 million times for training and optimizing their own models.