Anthropic recently removed tracking software used to monitor Chinese firms suspected of illicitly distilling its AI models to improve their own competing technology and bypass U.S. export controls.

Key Points

• Anthropic deployed and later removed spyware designed to identify Chinese companies using its Claude model to train their own AI systems.
• The company accused Alibaba’s Qwen team of using 25,000 fraudulent accounts to generate 28.8 million data exchanges for model distillation.
• Anthropic has banned approximately 700,000 accounts linked to China to prevent unauthorized intellectual property extraction.
• Similar accusations of model distillation have been directed at Chinese firms Deepseek, Moonshot, and MiniMax.
• A February 2025 study confirmed that most major Chinese AI models show evidence of distillation from U.S.-based systems.

Why it Matters

This conflict highlights a growing geopolitical struggle over AI supremacy, as U.S. firms argue that unauthorized distillation undermines American strategic and commercial advantages. The debate raises complex questions about intellectual property rights in an industry where model distillation is also a common, sometimes sanctioned, practice among domestic developers.