Anthropic Accuses Chinese Labs of AI Model Theft

On February 25, 2026, Anthropic publicly accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of executing large-scale distillation attacks on its Claude chatbot model. This technique allows attackers to bypass the substantial costs associated with training advanced AI models by effectively stealing intellectual property through high-volume API queries. The accused labs reportedly generated over 16 million exchanges with Claude, using a network of 24,000 fraudulent accounts to mimic legitimate user behavior, resulting in significant proprietary loss for Anthropic.

The implications of these accusations extend beyond a breach of intellectual property; they highlight vulnerabilities in the AI infrastructure that can threaten national security and competitive standing. The utilization of commercial proxy services, often linked to compromised IoT devices, underscores a growing trend where cybercriminal tactics are employed in AI development. This strategic move to institutionalize distillation attacks within parts of the Chinese tech sector might circumvent U.S. export controls on advanced semiconductor technologies, prompting a reassessment of data sovereignty and protective measures within the AI industry.