The 29-million-answer letter

Between April 22 and June 5, according to a letter Anthropic sent to senators and White House officials, about 25,000 fraudulent accounts ran some 28.8 million conversations with Claude, all aimed at its software-engineering and agentic-reasoning skills. Anthropic ties the operators to Alibaba's Qwen lab and calls it the largest distillation campaign it has ever caught — more answers than the three Chinese labs it accused in February (DeepSeek, MiniMax, Moonshot) pulled combined.

Nearly the same account count as February's three accused labs combined — but 1.8x the answers per account, as if the playbook is being tuned.

Distillation is the cheap copy: instead of training a rival model the hard way, you ask the expensive one millions of questions and train on its answers. The escalation here is twofold. It is the first time Anthropic has named a major conglomerate rather than a pure-play startup, and the campaign began after the White House science office had explicitly flagged the technique as a national-security concern.

The part to sit with is the sourcing. Every figure — the account count, the 29 million, the comparison — comes from one place: Anthropic's own letter. No outlet has independently verified them, and the method Anthropic used to attribute the accounts to Qwen is described nowhere. That same letter asks Congress for sanctions on distillers, tighter chip export controls, and an antitrust carve-out so frontier labs can share threat intelligence. Alibaba denies training on other models' outputs and declined to comment on the specifics.

So a company that stands to gain from sanctions and an antitrust exemption has supplied precise, alarming numbers that no one else can check, against a rival whose open-weight Qwen models are among the most downloaded on earth. The accusation may well be true. The point is that, for now, the only evidence is the word of the firm asking for the remedy.