... exchanges ~24,000 fraudulent accounts Proxy networks to bypass regional restrictions Targeted extraction of reasoning, coding and tool-use abilities Distillation is a common technique for training smaller models from larger ones. The controversy here is using a competitor s model outputs at scale without permission. What Anthropic Claims DeepSeek (~150k exchanges): Extracted step-by-step reasoning and censorship-safe reformulations. Moonshot (~3.4M exchanges): Targeted agentic reasoning, coding, tool use, and vision. MiniMax (~13M exchanges): Focused on agentic coding ... ... orchestration; allegedly pivoted quickly after a new Claude release. Anthropic says the activity showed identifiable patterns: repetitive, capability-focused prompts distributed across coordinated accounts ( hydra