Anthropic Accuses DeepSeek of AI Attacks

Anthropic Uncovers Large-Scale Distillation Attacks on Claude Models

• Anthropic identified illicit campaigns by DeepSeek, Moonshot AI, and MiniMax targeting its Claude models.
• The campaigns involved over 16 million exchanges via approximately 24,000 fraudulent accounts.
• DeepSeek conducted more than 150,000 exchanges focusing on reasoning tasks to generate training data.
• Moonshot AI executed over 3.4 million exchanges targeting agentic reasoning and coding capabilities.
• MiniMax was responsible for more than 13 million exchanges, detected during ongoing activity.

Anthropic has implemented new detection systems and strengthened account verification to counter these distillation attacks. The company emphasizes the need for coordinated action among AI labs, cloud providers, and policymakers to address such large-scale threats effectively.