Anthropic Says Its Defenses Are Reducing Claude Distillation Attack Effectiveness

Anthropic says its defenses are reducing the effectiveness of illicit distillation attacks on Claude. A February 2026 disclosure described over 16 million exchanges across roughly 24,000 accounts attributed to DeepSeek, Moonshot AI and MiniMax, while its September 2026 threat report cites campaigns tied to Alibaba, Moonshot and DeepSeek targeting Opus-class models.

• February 2026: over 16M exchanges across ~24,000 accounts
• Campaigns attributed to DeepSeek, Moonshot AI and MiniMax
• September 2026: Alibaba, Moonshot and DeepSeek hit Opus models
• Defenses: classifiers, coordinated-activity analysis, intel sharing