Anthropic Accuses Chinese AI Firms of Industrial-Scale IP Theft

AI developer Anthropic accused Chinese firms MiniMax, Moonshot AI, and DeepSeek of orchestrating industrial-scale distillation attacks to steal intellectual property from its Claude chatbot. Anthropic reported that these rivals used approximately 24,000 fraudulent accounts and proxy services to generate over 16 million exchanges, targeting agentic reasoning, coding, and tool use. MiniMax conducted the largest operation with 13 million interactions, followed by Moonshot AI with 3.4 million and DeepSeek with 150,000.

Anthropic and OpenAI warned that these activities pose national security risks, as distilled models may lack safety guardrails against the development of bioweapons or offensive cyber operations. Anthropic argued these attacks justify stricter U.S. export controls on AI chips, suggesting the scale of the theft indicates the firms possess advanced hardware. In response, Anthropic is implementing behavioral fingerprinting to identify statistical signatures of distillation queries.

Parallel to these accusations, Anthropic is embroiled in a dispute with the U.S. Department of Defense over a $200 million contract. The Pentagon threatened to ban the company after Anthropic maintained safety restrictions prohibiting the use of its AI for fully autonomous weapons or mass surveillance. Meanwhile, Elon Musk criticized Anthropic for allegedly stealing training data at a massive scale and accused its models of exhibiting demographic bias. Reports also indicate the U.S. military is deploying xAI's Grok model within classified systems amid the ongoing conflict with Anthropic.