Anthropic accuses DeepSeek, Moonshot, and MiniMax of industrial-scale model distillation via 24,000 fraudulent accounts
Feb 23, 2026
Key Points
- Anthropic accuses DeepSeek, Moonshot, and MiniMax of operating 24,000 fraudulent accounts to extract Claude's capabilities through 16 million queries for their own model training.
- The alleged distillation campaign represents industrial-scale IP theft rather than incidental usage, reversing the typical narrative where U.S. AI labs are accused of unauthorized training data extraction.
- The accusation exposes a competitive vulnerability: Chinese labs can systematically harvest frontier model outputs at scale without building foundational capabilities from scratch.
Summary
Anthropic publicly accused Chinese AI labs DeepSeek, Moonshot, and MiniMax of conducting industrial-scale distillation attacks on Claude. The companies created over 24,000 fraudulent accounts and generated more than 16 million exchanges with Claude to extract model capabilities for training their own systems.
Rather than building models from scratch, the labs allegedly reverse-engineered Anthropic's technology at scale by systematically querying Claude and harvesting outputs. The 16 million exchanges represent a methodical extraction campaign.
Social media reactions ranged from dark humor to pointed irony. One commenter noted the contradiction in IP enforcement, while another sarcastically questioned where respect for intellectual property had gone after Anthropic spent millions of man-hours on training data. The framing echoed earlier debates about training data sourcing, but inverted the usual dynamic with a U.S. AI leader accusing Chinese competitors of the extraction playbook.