China Tried to Clone Claude. It Leaked Their Own Secrets.
Anthropic's latest threat report reveals that seven Chinese labs, led by Alibaba, ran an industrial-scale distillation campaign against Claude, extracting chain-of-thought reasoning through 151 million queries from fake accounts. The report's core claim — that safety training is what makes a model expensive to copy — is supported by a detailed engineering story: one lab abandoned targeting a safer model for an older, less protected one. The timing against US export controls and Anthropic's own prior sanctions from the Pentagon for refusing military work complicates any simple 'theft' narrative, but the operational evidence is specific enough to demand attention.