Anthropic Accuses Chinese AI Firms of Large-Scale Model Distillation and IP Theft

Key Intelligence

• ■AI developer Anthropic alleges that Chinese firms, including DeepSeek and MiniMax, are engaging in "industrial-scale distillation attacks" against its Claude large language model.
• ■These firms reportedly used millions of queries (up to 16 million) to Claude via illicit or fake accounts to extract its capabilities and train their own rival AI models.
• ■The accusations echo similar concerns previously raised by OpenAI regarding Chinese AI companies exploiting proprietary models.
• ■Critics, including Elon Musk, have highlighted the perceived irony, suggesting Anthropic itself may have employed similar training methods in Claude's development.
• ■This incident, alongside reports of China's DeepSeek training AI models with advanced Nvidia chips despite US export bans, intensifies the debate around intellectual property rights, fair competition, and national security in the AI sector.