Anthropic has accused Alibaba's Qwen lab of operating an industrial-scale distillation campaign that extracted nearly 29 million responses from Claude over a three-month period using tens of thousands of fake accounts.
The allegations, detailed in a security report published Thursday, describe what Anthropic calls one of the largest known model-distillation operations targeting a Western AI system. According to the company, Alibaba-affiliated researchers created approximately 25,000 accounts on Claude's API platform and used automated tools to systematically query the model, capturing its outputs to train a competing system.
Anthropic said it detected the campaign through unusual traffic patterns that spiked after midnight Pacific Time, suggesting operators working from Asia. The queries reportedly focused on reasoning-intensive tasks including code generation, mathematical proofs, and multi-step logical analysis — areas where Claude has demonstrated particular strength.
Alibaba has not publicly responded to the allegations. The incident highlights growing tensions between Chinese and Western AI labs as competition intensifies and the boundaries between legitimate research and intellectual property extraction become increasingly blurred.
The dispute arrives as regulators in both Washington and Brussels consider rules governing model distillation and data scraping. Anthropic said it has strengthened its API rate limits and detection systems but warned that similar campaigns are likely underway against other frontier AI models.
Image source: i.ibb.co