Anthropic Says Chinese AI Firms Used 16 Million Claude Queries to Copy Model

Anthropic disclosed three industrial-scale distillation campaigns by DeepSeek, Moonshot AI, and MiniMax that used roughly 24,000 fraudulent accounts and commercial proxy “hydra” networks to generate over 16 million exchanges with the Claude model, extracting capabilities like agentic reasoning, tool use, coding, and censorship-evasion. Anthropic attributed the activity using request metadata and infrastructure indicators, implemented classifiers and behavioral fingerprinting to detect abuse, and warned that illicitly distilled models—lacking safeguards—pose serious national-security and misuse risks.