Anthropic accuses Alibaba of copying Claude
Anthropic told US senators that operators tied to Alibaba's Qwen lab used almost 25,000 fake accounts and 28.8M chats to extract Claude's coding skills.
Anthropic has accused Alibaba of running what it calls “the largest known distillation attack” on its Claude models. In a June 10 letter to the US Senate Banking Committee, reported on June 24, Anthropic said operators linked to Alibaba’s Qwen AI lab used almost 25,000 fraudulent accounts to generate more than 28.8 million exchanges with Claude between April 22 and June 5.
Distillation means using a strong model’s outputs to train a cheaper one. Anthropic says the accounts went after Claude’s most valuable skills, software engineering and agentic reasoning, and were set up to get around its rules that bar access from inside China. The letter went to committee chair Tim Scott and ranking member Elizabeth Warren, and asked for “coordinated action between government and industry.” Alibaba did not respond to requests for comment.
What it means for you
You will not feel this directly, but the knock-on effects reach every API customer. Expect providers to tighten identity checks, geographic limits, and rate caps to make bulk extraction harder, which adds verification friction for legitimate builders. It is also a reminder that part of why cheaper Chinese models and open-weights leaders like GLM-5.2 score close to the frontier can be imitation, not just independent training. Before you treat two similar benchmark numbers as proof that two models are interchangeable, it helps to understand what a model actually is and where its abilities came from.