Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek

Anthropic released a threat intelligence report on September 10, 2026 detailing five distillation attack campaigns by Chinese AI companies, totaling nearly 200 million exchanges. The largest campaign, attributed to Alibaba, involved 151 million exchanges across 3,500 accounts between May and July 2026, peaking at 3 million exchanges per day, targeting Claude's chain-of-thought reasoning to train Alibaba's Qwen models. A separate Moonshot AI campaign routed approximately 300,000 requests through 5,000 accounts over 10 days, with evidence of direct Chinese military involvement. Attackers used techniques such as disguising extraction queries as translation requests to bypass Anthropic's defenses and expose internal reasoning traces the company does not normally surface to users.