The old playbook was DDoS: flood a service with traffic to take it offline. The new one is far more dangerous. Sophisticated attackers are using thousands of fake or compromised accounts to query frontier AI models at massive scale, not to disrupt them, but to steal their intelligence. They harvest detailed chain-of-thought reasoning, tool use, and capability-specific outputs, then use it as training data to distill their own competing models. @AnthropicAI revealed this exact attack in Feb 2026: Chinese labs (@deepseek_ai , @moonshot , @MiniMax_AI ) generated over 16 million queries on @claudeai using ~24,000 fraudulent accounts.