Anthropic accuses Alibaba of largest-ever AI distillation attack on Claude
28.8 million fraudulent exchanges through 25,000 fake accounts extracted Claude's software-engineering and agentic-reasoning capabilities for Alibaba's Qwen lab
Add to a list
No lists yet.
Summary
Anthropic disclosed on June 24 that Alibaba's Qwen lab ran 28.8 million exchanges with Claude through roughly 25,000 fraudulent operator accounts between April 22 and June 5, 2026. The campaign targeted Claude's software engineering, agentic reasoning and long-horizon task capabilities, extracting them to train a rival model while stripping Claude's safety alignment. Anthropic's letter to Senators Tim Scott and Elizabeth Warren, sent June 10, called the campaign the largest known distillation attack on any US frontier lab, 80% larger by exchange volume than the combined DeepSeek-Moonshot-MiniMax cluster flagged in February 2026. Alibaba ignored a White House memo warning in April.
Why it matters
Distillation is legal arbitrage: no US statute explicitly bars querying a rival's API to train on the outputs, leaving Anthropic relying on terms-of-service enforcement. The case drives Congress toward codifying criminal penalties and gives the Trump administration a concrete predicate for tighter export controls on AI access.
What to watch
- Whether the Senate moves to legislate against distillation attacks following the letter.
- Alibaba or Qwen response and whether any model capability uplift is traceable to this campaign.
- Whether the FTC or DOJ open a formal investigation against Alibaba-affiliated entities in the US.