Anthropic says DeepSeek, Moonshot, and MiniMax used 24,000 fake accounts to rip off Claude
Anthropic says DeepSeek, Moonshot, and MiniMax used 24,000 fake accounts to rip off Claude
Publish Date: 2026-02-23 17:19:00
Source Domain: venturebeat.com
- Anthropic accused three Chinese AI labs of orchestrating elaborate, large-scale attempts to illegally extract capabilities from its models using fraudulent accounts.
- The labs targeted 16 million exchanges that violated Anthropic’s terms of service and access restrictions, focusing on extracting agentic reasoning, tool use, and coding capabilities.
- The techniques used likely involved mimicking clients and generating massive amounts of data to train rival models at a lower cost.
- Anthropic attributed large parts of the fraudulent activity to the labs using proxy networks that allowed access despite China bans on running the model.
- Anthropic framed the issue as a national security threat, arguing stolen capabilities could be used for offensive cyber operations and mass surveillance.
- Anthropic called for coordinated actions from industry, policymakers, and national security officials to combat the issue, highlighting the need for stricter export controls.
