Anthropic accuses DeepSeek, Moonshot, and MiniMax of industrial-scale distillation attacks on Claude

Anthropic publicly accused three Chinese AI labs of orchestrating coordinated campaigns to extract capabilities from Claude using 24,000 fake accounts and over 16 million exchanges. MiniMax ran the largest operation with 13M+ requests, pivoting to new Claude models within 24 hours of release. DeepSeek targeted reasoning chains and censorship-compliant answers, while Moonshot focused on agentic reasoning and tool use. The disclosure marks a dramatic escalation in US-China AI tensions as Washington debates chip export controls. Anthropic argued that illicitly distilled models strip safety guardrails, enabling potential military and surveillance misuse. OpenAI and Google have reported similar attacks.

View full digest for February 24, 2026