Anthropic Alleges Massive Intellectual Property Theft by Chinese AI Labs

Anthropic identified these activities as industrial-scale distillation, a practice where developers train smaller models using the outputs of more advanced AI. While distillation is common in legitimate research, Anthropic argues that this specific deployment effectively fast-tracks the development of military and surveillance tools by stripping away the safety guardrails embedded in the original model. The company warns that these unprotected capabilities are being redirected into offensive cyber operations and state-sponsored disinformation campaigns.

DeepSeek, which recently gained global attention for its high-efficiency architecture, is accused of conducting over 150,000 exchanges with Claude. Anthropic asserts that these queries were specifically designed to extract reasoning data and generate responses that circumvent political censorship regarding sensitive topics like state leadership and human rights. This development mirrors recent warnings from OpenAI, which also reported that DeepSeek has been actively attempting to free-ride on the proprietary advancements of American laboratories.