Anthropic Exposes Chinese AI Labs Running Massive Operation to Steal Claude's Capabilities Through 16 Million Illicit Exchanges
Summary
Anthropic exposes three Chinese AI labs — DeepSeek, Moonshot, and MiniMax — running an industrial-scale operation using 24,000 fraudulent accounts to conduct over 16 million illicit exchanges, stealing Claude's capabilities while bypassing critical safety safeguards designed to prevent misuse for bioweapons and cyberattacks.
Key Points
- Anthropic identifies three Chinese AI labs — DeepSeek, Moonshot, and MiniMax — conducting 'industrial-scale' model distillation attacks, using approximately 24,000 fraudulent accounts to generate over 16 million illicit exchanges aimed at stealing Claude's capabilities.
- The unauthorized distillation bypasses critical safety safeguards that prevent AI models from being used for dangerous purposes such as bioweapon development or cyberattacks, with risks compounding further if the distilled models are open-sourced.
- Anthropic joins OpenAI and Google in sounding the alarm on model distillation attacks this month, calling for rapid, coordinated action across the AI industry, policymakers, and the global community, warning that 'no company can solve this alone.'