Anthropic accuses Alibaba of 28.8 million-query distillation attack on Claude, targeting agentic reasoning

In a June 10 letter to the Senate Banking Committee, Anthropic alleged operators tied to Alibaba's Qwen lab used ~25,000 fraudulent accounts to harvest Claude's agentic reasoning and software-engineering capabilities — the largest known distillation campaign against the company.

Anthropic told the Senate Banking Committee that operators tied to Alibaba’s Qwen AI lab ran 28.8 million unauthorized exchanges with Claude between April 22 and June 5, using roughly 25,000 fraudulent accounts to harvest agentic reasoning and software-engineering capabilities. The letter, dated June 10 and addressed to Chair Tim Scott and Ranking Member Elizabeth Warren, was first reported by Bloomberg on June 24 and quickly obtained by Reuters, CNBC, and Nikkei Asia. Anthropic calls it the largest known distillation campaign ever run against the company.

The scale is the story. In February, Anthropic disclosed three earlier campaigns it attributed to DeepSeek (more than 150,000 exchanges), Moonshot AI (over 3.4 million), and MiniMax (over 13 million). The Qwen-linked operation is larger than all three combined, by an order of magnitude on the smaller ones. It also targeted a specific layer of the stack: not raw text generation, but the agentic and coding behaviors that frontier labs treat as their actual moat.

The timing is the subtext. On June 8, the Pentagon added Alibaba to its Chinese military companies list. On June 12, the Commerce Department imposed export restrictions on Anthropic’s Fable 5 and Mythos 5 models. This week, Alibaba sued the Defense Department to remove itself from the list. The June 10 letter, sitting between those dates, isn’t just a security disclosure. It’s a positioning document for senators including Bill Hagerty and Andy Kim who are actively writing the rules for what counts as a national-security-relevant AI export.

Alibaba’s American depositary receipts fell more than three percent on the news, dropping below $100.

There’s a recognizable shape here, last seen during the late-2010s Huawei cycle: a US company arrives on Capitol Hill with a granular technical narrative just as the regulatory perimeter is being redrawn around its Chinese competitor. Whether the 28.8 million queries actually trained a Qwen model is, for the policy audience, almost beside the point. The number is the artifact. The letter is the artifact. The committee now owns the framing.

Anthropic accuses Alibaba of 28.8 million-query distillation attack on Claude, targeting agentic reasoning

Sources