This isn’t a small incident. Anthropic has formally accused Alibaba of waging a large-scale campaign to illegally distill Claude’s capabilities, naming the Chinese tech giant in a letter sent to the US Senate Banking Committee and White House officials. It’s the first time Anthropic has pointed the finger at a major Chinese conglomerate — previous accusations targeted smaller startups like DeepSeek, MiniMax, and Moonshot AI.
What happened
Between April 22 and June 5, 2026, operators linked to Alibaba’s Qwen AI lab used nearly 25,000 fraudulent accounts to generate 28.8 million exchanges with Claude. The targets were specific: software engineering and agentic reasoning — Claude’s most commercially valuable skills.
The technique is called ‘adversarial distillation’. You systematically prompt an advanced model to harvest its reasoning patterns and data structures, then use those responses to train your own models. It bypasses millions of dollars in R&D costs by effectively copying someone else’s homework at scale.
Why it matters
The accusation lands at a moment when tensions between US AI companies and China are already at a peak. Fable 5 and Mythos 5 have been offline since June 12 following a White House export control directive. Anthropic is now essentially providing evidence for why such controls are necessary — at least from their perspective.
For Alibaba and the Qwen team, the allegation is awkward. Qwen has been one of China’s strongest open-source models, improving rapidly in recent months. The question of how much of that progress comes from original training versus ‘borrowed’ capabilities will now be asked much more loudly.
My take
Distillation isn’t new in the AI world — many labs do it in some form. But 25,000 fake accounts and nearly 29 million queries over six weeks is a different scale entirely. This isn’t a gray area anymore — it’s industrial-scale IP extraction using AI tools.
At the same time, Anthropic clearly has an interest in making this public right now. While negotiations with the White House over restoring Fable 5 access are ongoing, it doesn’t hurt to show senators exactly why American AI models need protection from Chinese access.
The truth probably lies somewhere in between. But the numbers are both impressive and unsettling.
Sources: