Anthropic’s breakdown shows different objectives across the campaigns. DeepSeek reportedly sent more than 150,000 requests aimed at eliciting chain-of-thought style reasoning, reward-model data for reinforcement learning, and censorship-compliant answers on politically sensitive topics. Moonshot AI allegedly generated more than 3.4 million requests focused on agentic reasoning, tool use, programming, data analysis, computer vision, and reconstruction of Claude’s reasoning patterns. MiniMax, with more than 13 million requests, was described as the largest campaign, centered on agentic coding and tool orchestration.

Anthropic says MiniMax shifted rapidly when a new Claude model launched, redirecting nearly half of its traffic to the new system within 24 hours. The company is calling for a coordinated response from both industry and policymakers. Anthropic also notes that OpenAI and Google have reported similar attempts linked to Chinese actors.